Slide plate and method for dynamically switching between adaptive connection and connection optimization through machine learning

Abstract

The present invention relates to technology for machine learning solutions in dynamic switching between adaptive connections and connection optimization. The technology for adapting a communication protocol (e.g., TCP / IP, UDP, etc.) to network communications between endpoints (e.g., accelerated cores configured within an accelerator device) includes a sled having a computing engine. The computing engine monitors telemetry data associated with one or more network communications between a given core and another core. The network communications are established via a given communication protocol. Based on the monitored telemetry data, the computing engine determines that a condition has been triggered to change the network communications from the given communication protocol to another communication protocol. The computing engine then converts the network communications to the other communication protocol.

Description

[0001] Cross-reference to related applications

[0002] This application claims the benefit of Indian Provisional Patent Application No. 201741030632, filed on August 30, 2017, and U.S. Provisional Patent Application No. 62 / 584,401, filed on November 10, 2017. Background Art

[0003] In a system that distributes workloads across multiple computing devices (e.g., in a data center), centralized servers can form nodes of computing devices used to process the workload. Each node represents a logical aggregation of resources (e.g., computing, storage, acceleration, etc.) provided by each computing device. For example, a node may include a computing device configured with a hardware accelerator, such as a field programmable gate array (FPGA) device and / or a graphics processing unit (GPU). Generally, hardware accelerators improve the execution speed of workload functions. To accelerate a given function of a workload (such as a given function of an application), the centralized server can configure the accelerator device with an accelerated kernel suitable for accelerating the task.

[0004] Once completed, the accelerator device returns the data generated by the accelerated function to the application. In some cases, the system can provide a core-to-core network, which allows a given core to transmit the resulting data to another core for further processing of the workload. The cores in the accelerator device can establish network communication with other core devices via some network communication protocol, such as TCP / IP (Transmission Control Protocol / Internet Protocol) or UDP (User Datagram Protocol). Based on the current resource load in the system, using a particular network communication protocol may be more efficient than using another protocol. BRIEF DESCRIPTION OF THE DRAWINGS

[0005] In the accompanying figures, the concepts described herein are illustrated by way of example and not by way of limitation. For simplicity and clarity of illustration, the elements illustrated in the figures are not necessarily drawn to scale. Where deemed appropriate, reference labels have been repeated between the figures to indicate corresponding or similar elements.

[0006] Figure 1 is a simplified diagram of at least one embodiment of a data center for utilizing disaggregated resources to execute workloads;

[0007] Figure 2 yes Figure 1 A simplified diagram of at least one embodiment of a pod in a data center;

[0008] Figure 3 can be included in Figure 2A perspective view of at least one embodiment of a rack in a pod;

[0009] Figure 4 yes Figure 3 A side elevation view of a rack;

[0010] Figure 5 It is equipped with a sled. Figure 3 A perspective view of the rack;

[0011] Figure 6 yes Figure 5 A simplified block diagram of at least one embodiment of a top side of a slide;

[0012] Figure 7 yes Figure 6 A simplified block diagram of at least one embodiment of a bottom side of a slide;

[0013] Figure 8 is Figure 1 A simplified block diagram of at least one embodiment of a computing sled that may be used in a data center;

[0014] Figure 9 yes Figure 8 a top perspective view of at least one embodiment of a computing sled;

[0015] Figure 10 is Figure 1 A simplified block diagram of at least one embodiment of an accelerator sled that may be used in a data center;

[0016] Figure 11 yes Figure 10 A top perspective view of at least one embodiment of an accelerator slide;

[0017] Figure 12 is Figure 1 A simplified block diagram of at least one embodiment of a storage sled that may be used in a data center;

[0018] Figure 13 yes Figure 12 a top perspective view of at least one embodiment of a storage slide;

[0019] Figure 14 is Figure 1 A simplified block diagram of at least one embodiment of a memory sled that may be used in a data center; and

[0020] Figure 15 Yes, you can Figure 1 A simplified block diagram of a system established within a data center to execute workloads using managed nodes composed of disaggregated resources;

[0021] Figure 16 is a simplified block diagram of at least one embodiment of a system for adapting a communication protocol to network communications between endpoints;

[0022] Figure 17 yes Figure 16 A simplified block diagram of at least one embodiment of an accelerator skateboard of a system;

[0023] Figure 18 It can be done through Figure 16 and 17 A simplified block diagram of at least one embodiment of an environment established by an accelerator skateboard;

[0024] 19A and 19B are diagrams of example embodiments of a core-to-core communication network; and

[0025] Figure 20 and 21 is a simplified flow chart of at least one embodiment of a method for adapting a communication protocol to network communications between endpoints. DETAILED DESCRIPTION

[0026] While the concepts of the present disclosure are susceptible to various modifications and alternative forms, specific embodiments thereof have been shown by way of example in the drawings and will be described in detail herein. However, it should be understood that there is no intention to limit the concepts of the present disclosure to the particular forms disclosed, but on the contrary, the present disclosure is intended to cover all modifications, equivalents, and alternatives consistent with the present disclosure and the appended claims.

[0027] References in the specification to "one embodiment," "an embodiment," "illustrative embodiment," etc., indicate that the described embodiment may include a particular feature, structure, or characteristic, but every embodiment may or may not include that particular feature, structure, or characteristic. Furthermore, such phrases do not necessarily refer to the same embodiment. Furthermore, when a particular feature, structure, or characteristic is described in conjunction with an embodiment, it is understood that implementation of such feature, structure, or characteristic in conjunction with other embodiments is within the knowledge of those skilled in the art, regardless of whether explicitly described. Additionally, it should be understood that items included in a list of the form "at least one of A, B, and C" may mean (A); (B); (C); (A and B); (A and C); (B and C); or (A, B, and C). Similarly, items listed in the form "at least one of A, B, or C" may mean (A); (B); (C); (A and B); (A and C); (B and C); or (A, B, and C).

[0028] In some cases, the disclosed embodiments may be implemented in hardware, firmware, software, or any combination thereof. The disclosed embodiments may also be implemented as instructions carried by or stored on a transitory or non-transitory machine-readable (e.g., computer-readable) storage medium, which instructions may be read and executed by one or more processors. A machine-readable storage medium may be embodied as any storage device, mechanism, or other physical structure (e.g., volatile or non-volatile memory, media disk, or other media device) for storing or transmitting information in a form readable by a machine.

[0029] In the accompanying drawings, some structural or method features may be shown in a particular arrangement and / or order. However, it should be understood that such a particular arrangement and / or order may not be required. On the contrary, in some embodiments, such features may be arranged in a manner and / or order different from that shown in the illustrative figures. Additionally, the inclusion of structural or method features in a particular figure is not intended to imply that such features are required in all embodiments, and in some embodiments, such features may not be included or may be combined with other features.

[0030] Now refer to Figure 1A data center 100, in which disaggregated resources can collaboratively execute one or more workloads (e.g., collaboratively execute applications on behalf of customers), includes multiple pods 110, 120, 130, and 140, each of which includes one or more rows of racks. As described in greater detail herein, each rack houses multiple sleds, each of which can be embodied as a computing device, such as a server, primarily equipped with a specific type of resource (e.g., memory devices, data storage devices, accelerator devices, general-purpose processors). In an illustrative embodiment, the sleds in each pod 110, 120, 130, and 140 are connected to multiple pod switches (e.g., switches that route data communications to and from the sleds within the pod). The pod switches, in turn, are connected to a spine switch 150, which switches communications between the pods (e.g., pods 110, 120, 130, and 140) in the data center 100. In some embodiments, the sleds can be connected to a fabric using Intel Omni-Path technology. As described in greater detail herein, resources within sleds in data center 100 can be assigned to groups (referred to herein as "managed nodes") that contain resources from one or more other sleds to be collectively utilized in executing workloads. Workloads can be executed as if the resources belonging to the managed nodes were located on the same sled. Resources in the managed nodes can even belong to sleds belonging to different racks and even to different pods 110, 120, 130, 140. Some resources of a single sled can be assigned to one managed node, while other resources of the same sled can be assigned to a different managed node (e.g., one processor can be assigned to one managed node, and another processor from the same sled can be assigned to a different managed node). By disaggregating resources into sleds that predominantly include a single type of resource (e.g., a compute sled that primarily includes compute resources, a memory sled that primarily includes memory resources), and selectively allocating and de-allocating the disaggregated resources to form managed nodes assigned to execute workloads, data center 100 provides more efficient resource utilization than a typical data center that includes hyperconverged servers that include compute, memory, storage, and perhaps additional resources. As such, data center 100 can provide better performance (eg, throughput, operations per second, latency, etc.) than a typical data center with the same number of resources.

[0031] Now refer to Figure 2In the illustrated embodiment, pod 110 includes a collection of rows 200, 210, 220, 230 of racks 240. Each rack 240 can accommodate multiple sleds (e.g., sixteen sleds) and provide power and data connectivity to the accommodated sleds, as described in greater detail herein. In the illustrated embodiment, the racks in each row 200, 210, 220, 230 are connected to a plurality of pod switches 250, 260. Pod switches 250 include a set of ports 252 to which the sleds of the racks of pod 110 are connected, and another set of ports 254 that connect pod 110 to spine switch 150 to provide connectivity to other pods in data center 100. Similarly, pod switch 260 includes a set of ports 262 to which the sleds of the racks of pod 110 are connected, and a set of ports 264 that connect pod 110 to spine switch 150. As such, the use of this pair of switches 250, 260 provides a level of redundancy for pod 110. For example, if either of the switches 250, 260 fails, the sleds in the pod 110 can still maintain data communications with the rest of the data center 100 (e.g., sleds of other pods) through the other switch 250, 260. Furthermore, in the illustrative embodiment, the switches 150, 250, 260 may embody dual-mode optical switches capable of routing both Ethernet protocol communications carrying Internet Protocol (IP) packets and communications according to a second, high-performance link layer protocol (e.g., Infiniband of Intel's Omni-Path Architecture) via the optical signaling medium of fiber.

[0032] It should be appreciated that each of the other pods 120, 130, 140 (and any additional pods of the data center 100) can be similarly configured to operate in a manner similar to that of the server. Figure 2 Shown in and about Figure 2 pod 110 is depicted and has similar components to pod 110 (e.g., each pod may have multiple rows of racks housing multiple sleds, as described above). Additionally, while two pod switches 250, 260 are shown, it should be understood that in other embodiments, each pod 110, 120, 130, 140 may be connected to a different number of pod switches (e.g., to provide even greater failover capability).

[0033] Now refer to Figure 3-5, each illustrative rack 240 of the data center 100 includes two vertically arranged elongated support columns 302, 304. For example, the elongated support columns 302, 304 extend upward from the floor of the data center 100 when deployed. The rack 240 also includes one or more horizontal pairs 310 (at the top) of elongated support arms 312 configured to support the slides of the data center 100. Figure 3 One of the pair of elongated support arms 312 extends outwardly from the elongated support column 302 and the other elongated support arm 312 extends outwardly from the elongated support column 304.

[0034] In the illustrative embodiment, each sled of the data center 100 is embodied as a chassis-less sled. That is, each sled has a chassis-less circuit board substrate on which physical resources (e.g., processors, memory, accelerators, storage, etc.) are mounted, as discussed in greater detail below. As such, the rack 240 is configured to accommodate chassis-less sleds. For example, each pair 310 of elongated support arms 312 defines a sled slot 320 of the rack 240 that is configured to accommodate a corresponding chassis-less sled. To do so, each illustrative elongated support arm 312 includes a circuit board guide 330 configured to accommodate the sled's chassis-less circuit board substrate. Each circuit board guide 330 is secured to or otherwise mounted to a front face 332 of a corresponding elongated support arm 312. For example, in the illustrative embodiment, each circuit board guide 330 is mounted at a distal end of the corresponding elongated support arm 312 relative to the corresponding elongated support column 302, 304. For clarity of the figures, not every circuit board guide 330 may be referenced in every figure.

[0035] Each circuit board guide 330 includes inner walls defining a circuit board slot 380 configured to receive the chassis-less circuit board substrate of the sled 400 when the sled 400 is received in the corresponding sled slot 320 of the chassis 240. To do so, Figure 4 As shown in FIG, a user (or robot) aligns the chassis-less circuit board substrate of the illustrative chassis-less skateboard 400 with the skateboard slot 320. The user or robot can then slide the chassis-less circuit board substrate forward into the skateboard slot 320 such that each side edge 414 of the chassis-less circuit board substrate is received in a corresponding circuit board slot 380 of the pair 310 of elongated support arms 312 defining the corresponding skateboard slot 320, as shown in FIG. Figure 4. By having a robotically accessible and robotically manipulable sled that includes disassembled resources, each type of resource can be upgraded independently of each other and at their own optimized refresh rate. In addition, the sled is configured to blind-mate with the power and data communication cables in each rack 240, thereby enhancing its ability to be quickly removed, upgraded, reinstalled and / or replaced. As such, in some embodiments, the data center 100 can operate on the floor of the data center (e.g., perform workloads, undergo maintenance and / or upgrades, etc.) without human involvement. In other embodiments, a human can facilitate one or more maintenance or upgrade operations in the data center 100.

[0036] It should be appreciated that each circuit board guide 330 is double-sided. That is, each circuit board guide 330 includes inner walls that define a circuit board slot 380 on each side of the circuit board guide 330. In this manner, each circuit board guide 330 can support a chassis-less circuit board substrate on either side. As such, a single additional extended support column can be added to the rack 240 to transform the rack 240 into one that can hold a plurality of circuit boards. Figure 3 . The illustrative rack 240 includes seven pairs 310 of elongated support arms 312 defining seven corresponding sled slots 320, each of which is configured to accommodate and support a corresponding sled 400 as discussed above. Of course, in other embodiments, the rack 240 may include additional or fewer pairs 310 of elongated support arms 312 (i.e., additional or fewer sled slots 320). It should be appreciated that because the sled 400 is chassisless, it can have an overall height different from that of a typical server. Accordingly, in some embodiments, the height of each sled slot 320 can be shorter than the height of a typical server (e.g., shorter than a single rack unit (1U)). That is, the vertical distance between each pair 310 of elongated support arms 312 can be less than a standard rack unit (1U). Additionally, due to the relative reduction in the height of the sled slots 320, the overall height of the rack 240 can, in some embodiments, be shorter than that of a conventional rack enclosure. For example, in some embodiments, each of the extended support columns 302, 304 can have a length of six feet or less. Again, in other embodiments, the rack 240 can have different dimensions. Further, it should be appreciated that the rack 240 does not include any walls, enclosures, or the like. Rather, the rack 240 is an enclosure-less rack that is open to the local environment. Of course, in some cases, such as those where the rack 240 forms the end-of-row rack in the data center 100, an end plate can be attached to one of the extended support columns 302, 304.

[0037] In some embodiments, various interconnects can be routed upward or downward through the elongated support columns 302, 304. To facilitate such routing, each elongated support column 302, 304 includes inner walls defining an inner cavity in which the interconnects can be located. The interconnects routed through the elongated support columns 304, 304 can embody any type of interconnect, including but not limited to data or communication interconnects that provide a communication connection to each sled slot 320, power interconnects that provide power to each sled slot 320, and / or other types of interconnects.

[0038] In the illustrative embodiment, the rack 240 includes a support platform on which corresponding optical data connectors (not shown) are mounted. Each optical data connector is associated with a corresponding sled slot 320 and is configured to mate with the optical data connector of the corresponding sled 400 when the sled 400 is received in the corresponding sled slot 320. In some embodiments, the optical connections between the components in the data center 100 (e.g., sleds, racks, and switches) are made using blind-mate optical connections. For example, a door on each cable can prevent dust from contaminating the optical fibers within the cable. During connection to the blind-mate optical connector mechanism, the door is pushed open as the end of the cable enters the connector mechanism. Subsequently, the optical fibers within the cable enter the gel within the connector mechanism, and the optical fibers of one cable come into contact with the optical fibers of another cable within the gel within the connector mechanism.

[0039] The illustrative rack 240 also includes a fan array 370 coupled to the cross support arms of the rack 240. The fan array 370 includes one or more rows of cooling fans 372 aligned horizontally between the elongated support columns 302, 304. In the illustrative embodiment, the fan array 370 includes a row of cooling fans 372 for each sled slot 320 of the rack 240. As discussed above, in the illustrative embodiment, each sled 400 does not include any onboard cooling system, and as such, the fan array 370 provides cooling for each sled 400 received in the rack 240. In the illustrative embodiment, each rack 240 also includes a power supply associated with each sled slot 320. Each power supply is secured to one of the elongated support arms 312 in the pair 310 of elongated support arms 312 that define the corresponding sled slot 320. For example, the rack 240 may include a power supply coupled to or secured to each elongated support arm 312 extending from the elongated support column 302. Each power supply includes a power connector that is configured to mate with a power connector of a sled 400 when the sled 400 is received in the corresponding sled slot 320. In the illustrative embodiment, the sleds 400 do not include any onboard power supplies, and as such, a power supply provided in the rack 240 supplies power to the corresponding sled 400 when mounted to the rack 240.

[0040] Now refer to Figure 6 In the illustrative embodiment, the sleds 400 are configured to be installed in corresponding racks 240 of the data center 100, as discussed above. In some embodiments, each sled 400 may be optimized or otherwise configured to perform a specific task, such as a computing task, an acceleration task, a data storage task, etc. For example, the sled 400 may be embodied as described below with respect to Figure 8-9 The computational sled 800 discussed below with respect to Figure 10-11 The accelerator slide 1000 discussed below with respect to Figure 12-13 The storage sled 1200 discussed above may also be embodied as a sled optimized or otherwise configured to perform other specialized tasks, such as those discussed below with respect to Figure 14 Memory sled 1400 is discussed.

[0041] As discussed above, the illustrative sled 400 includes a chassis-less circuit board substrate 602 that supports various physical resources (e.g., electrical components) mounted thereon. It should be appreciated that the circuit board substrate 602 is "chassis-less" in that the sled 400 does not include a housing or enclosure. Instead, the chassis-less circuit board substrate 602 is open to the local environment. The chassis-less circuit board substrate 602 can be composed of any material capable of supporting the various electrical components mounted thereon. For example, in the illustrative embodiment, the chassis-less circuit board substrate 602 is composed of an FR-4 glass-reinforced epoxy laminate. Of course, in other embodiments, other materials can be used to form the chassis-less circuit board substrate 602.

[0042] As discussed in greater detail below, chassisless circuit board substrate 602 includes several features that improve the thermal cooling characteristics of the various electrical components mounted on chassisless circuit board substrate 602. As discussed, chassisless circuit board substrate 602 does not include a housing or enclosure, which can improve airflow over the electrical components of sled 400 by reducing structures that could inhibit airflow. For example, because chassisless circuit board substrate 602 is not positioned in a separate housing or enclosure, there is no base plate (e.g., a back plate of a chassis) to chassisless circuit board substrate 602, which could inhibit airflow across the electrical components. Additionally, chassisless circuit board substrate 602 has a geometry configured to reduce the length of the airflow path across the electrical components mounted to chassisless circuit board substrate 602. For example, the illustrative chassisless circuit board substrate 602 has a width 604 that is greater than a depth 606 of chassisless circuit board substrate 602. In one particular embodiment, for example, the chassisless circuit board substrate 602 has a width of approximately 21 inches and a depth of approximately 9 inches, compared to a typical server having a width of approximately 17 inches and a depth of approximately 39 inches. As such, the airflow path 608 extending from the front edge 610 toward the rear edge 612 of the chassisless circuit board substrate 602 has a shorter distance relative to a typical server, which can improve the thermal cooling characteristics of the sled 400. Additionally, while not present in the FIG. Figure 6 6. However, the various physical resources mounted to the chassisless circuit board substrate 602 are mounted in corresponding positions so that two substantially heat-generating electrical components do not shield each other, as discussed in greater detail below. That is, two electrical components that generate a considerable amount of heat during operation (i.e., greater than a nominal amount of heat sufficient to adversely affect the cooling of another electrical component) are not mounted to the chassisless circuit board substrate 602 in a straight line with each other along the direction of the airflow path 608 (i.e., along a direction extending from the front edge 610 toward the rear edge 612 of the chassisless circuit board substrate 602).

[0043] As discussed above, the illustrative sled 400 includes one or more physical resources 620 mounted to the front side 650 of the chassis-less circuit board substrate 602. Figure 6, two physical resources 620 are shown, but it should be appreciated that in other embodiments the sled 400 may include one, two, or more physical resources 620. Depending on, for example, the type or intended functionality of the sled 400, the physical resources 620 may be embodied as any type of processor, controller, or other computing circuitry capable of performing various tasks (such as computing functions) and / or controlling the functionality of the sled 400. For example, as discussed in greater detail below, the physical resources 620 may be embodied as a high-performance processor in embodiments where the sled 400 is embodied as a computing sled, as an accelerator coprocessor or circuit in embodiments where the sled 400 is embodied as an accelerator sled, as a memory controller in embodiments where the sled 400 is embodied as a storage sled, or as a set of memory devices in embodiments where the sled 400 is embodied as a memory sled.

[0044] The sled 400 also includes one or more additional physical resources 630 mounted to the front side 650 of the chassisless circuit board substrate 602. In the illustrative embodiment, the additional physical resources include a network interface controller (NIC), as discussed in greater detail below. Of course, depending on the type and functionality of the sled 400, in other embodiments the physical resources 630 may include additional or other electrical components, circuits, and / or devices.

[0045] Physical resource 620 is communicatively coupled to physical resource 630 via input / output (I / O) subsystem 622. I / O subsystem 622 may be embodied as circuitry and / or components used to facilitate input / output operations with physical resource 620, physical resource 630, and / or other components of sled 400. For example, I / O subsystem 622 may be embodied as or otherwise include a memory controller hub, an input / output control hub, an integrated sensor hub, a firmware device, communication links (e.g., point-to-point links, bus links, wires, cables, optical guides, printed circuit board traces, etc.), and / or other components and subsystems used to facilitate input / output operations. In an illustrative embodiment, I / O subsystem 622 may be embodied as or otherwise include a double data rate 4 (DDR4) data bus or a DDR5 data bus.

[0046] In some embodiments, sled 400 may also include a resource-to-resource interconnect 624. Resource-to-resource interconnect 624 may be embodied as any type of communication interconnect capable of facilitating resource-to-resource communication. In an illustrative embodiment, resource-to-resource interconnect 624 is embodied as a high-speed point-to-point interconnect (e.g., faster than I / O subsystem 622). For example, resource-to-resource interconnect 624 may be embodied as a Quick Path Interconnect (QPI), an Ultra Path Interconnect (UPI), or other high-speed point-to-point interconnect dedicated for resource-to-resource communication.

[0047] The sled 400 also includes a power connector 640 configured to mate with a corresponding power connector of the rack 240 when the sled 400 is installed in the corresponding rack 240. The sled 400 receives power from the rack 240's power supply via the power connector 640 to supply power to the various electrical components of the sled 400. That is, the sled 400 does not include any local power supply (i.e., an onboard power supply) for providing power to the electrical components of the sled 400. The elimination of a local or onboard power supply facilitates a reduction in the overall footprint of the chassisless circuit board substrate 602, which can enhance thermal cooling characteristics of the various electrical components mounted on the chassisless circuit board substrate 602, as discussed above. In some embodiments, power is provided to the processor 820 via vias directly beneath the processor 820 (e.g., through the back surface 750 of the chassisless circuit board substrate 602), thereby providing an increased thermal budget, additional current and / or voltage, and better voltage control than a typical board.

[0048] In some embodiments, the sled 400 may further include mounting features 642 configured to cooperate with a mounting arm or other structure of the robot to facilitate placement of the sled 600 in the rack 240 by the robot. The mounting features 642 may be embodied as any type of physical structure that allows the robot to grasp the sled 400 without damaging the chassisless circuit board substrate 602 or the electrical components mounted thereto. For example, in some embodiments, the mounting features 642 may be embodied as non-conductive pads that attach to the chassisless circuit board substrate 602. In other embodiments, the mounting features may be embodied as brackets, posts, or other similar structures that attach to the chassisless circuit board substrate 602. The specific number, shape, size, and / or composition of the mounting features 642 may depend on the design of the robot configured to manage the sled 400.

[0049] Now refer to Figure 7 In addition to the physical resources 630 mounted on the front side 650 of the chassisless circuit board substrate 602, the sled 400 also includes one or more memory devices 720 mounted to the back side 750 of the chassisless circuit board substrate 602. That is, the chassisless circuit board substrate 602 is embodied as a double-sided circuit board. The physical resources 620 are communicatively coupled to the memory devices 720 via the I / O subsystem 622. For example, the physical resources 620 and the memory devices 720 can be communicatively coupled via one or more through-holes extending through the chassisless circuit board substrate 602. In some embodiments, each physical resource 620 can be communicatively coupled to a different set of one or more memory devices 720. Alternatively, in other embodiments, each physical resource 620 can be communicatively coupled to each memory device 720.

[0050] The memory device 720 may be embodied as any type of memory device capable of storing data for the physical resources 620 during operation of the sled 400, such as any type of volatile memory (e.g., dynamic random access memory (DRAM), etc.) or non-volatile memory. Volatile memory may be a storage medium that requires power to maintain the state of the data stored by the medium. Non-limiting examples of volatile memory may include various types of random access memory (RAM), such as dynamic random access memory (DRAM) or static random access memory (SRAM). One specific type of DRAM that may be used in the memory module is synchronous dynamic random access memory (SDRAM). In certain embodiments, the DRAM of the memory component may comply with standards promulgated by JEDEC, such as JESD79F for DDR SDRAM, JESD79-2F for DDR2 SDRAM, JESD79-3F for DDR3 SDRAM, JESD79-4A for DDR4 SDRAM, JESD209 for Low Power DDR (LPDDR), JESD209-2 for LPDDR2, JESD209-3 for LPDDR3, and JESD209-4 for LPDDR4 (these standards are available at www.jedec.org). Such standards (and similar standards) may be referred to as DDR-based standards, and the communication interface of a memory device that implements such standards may be referred to as a DDR-based interface.

[0051] In one embodiment, the memory device is a block addressable memory device, such as those based on NAND or NOR technology. The memory device may also include next generation non-volatile devices, such as Intel 3D XPoint TMMemory or other types of addressable, write-in-place non-volatile memory devices. In one embodiment, the memory device may be or may include a memory device using chalcogenide glass, multi-threshold level NAND flash memory, NOR flash memory, single or multi-level phase change memory (PCM), resistive memory, nanowire memory, ferroelectric transistor random access memory (FeTRAM), antiferroelectric memory, magnetoresistive random access memory (MRAM) memory incorporating memristor technology, resistive memory including metal oxide substrate, oxygen vacancy substrate, and conductive bridge random access memory (CB-RAM), or spin transfer torque (STT)-MRAM, devices based on spintronic magnetic junction memory, devices based on magnetic tunnel junction (MTJ), devices based on DW (domain wall) and SOT (spin-orbit transfer), memory devices based on thyristors, or any combination of the above memory devices, or other memory devices. The memory device may refer to the die itself and / or the packaged memory product. In some embodiments, a memory device may include a transistor-free stackable cross-point architecture in which memory cells are located at the intersection of word lines and bit lines and are individually addressable and in which bit storage is based on changes in bulk resistance.

[0052] Now refer to Figure 8 In some embodiments, the sled 400 may be embodied as a computing sled 800. The computing sled 800 is optimized or otherwise configured to perform computing tasks. Of course, as discussed above, the computing sled 800 may rely on other sleds, such as an acceleration sled and / or a storage sled, to perform such computing tasks. The computing sled 800 includes various physical resources (e.g., electrical components) similar to those of the sled 400, which have been Figure 8 The same reference numbers are used to identify the above Figure 6 and 7 The descriptions provided for such components are applicable to corresponding components of the computing sled 800 and are not repeated herein for clarity of the description of the computing sled 800 .

[0053] In the illustrative computing sled 800, the physical resource 620 is embodied as a processor 820. Figure 8Only two processors 820 are shown in the figure, but it should be appreciated that in other embodiments the computing sled 800 may include additional processors 820. Illustratively, the processors 820 are embodied as high-performance processors 820 and may be configured to operate at a relatively high power rating. Although the processors 820 operating at a greater power rating than typical processors (which operate at approximately 155-230W) generate additional heat, the enhanced thermal cooling characteristics of the chassis-less circuit board substrate 602 discussed above facilitate higher power operation. For example, in the illustrative embodiment, the processors 820 are configured to operate at a power rating of at least 250W. In some embodiments, the processors 820 may be configured to operate at a power rating of at least 350W.

[0054] In some embodiments, the compute sled 800 may also include a processor-to-processor interconnect 842. Similar to the resource-to-resource interconnect 624 of the sled 400 discussed above, the processor-to-processor interconnect 842 may be embodied as any type of communication interconnect capable of facilitating communication across the processor-to-processor interconnect 842. In an illustrative embodiment, the processor-to-processor interconnect 842 is embodied as a high-speed point-to-point interconnect (e.g., faster than the I / O subsystem 622). For example, the processor-to-processor interconnect 842 may be embodied as a Quick Path Interconnect (QPI), an Ultra Path Interconnect (UPI), or other high-speed point-to-point interconnect dedicated for processor-to-processor communication.

[0055] The computing sled 800 also includes communication circuitry 830. The illustrative communication circuitry 830 includes a network interface controller (NIC) 832, which may also be referred to as a host fabric interface (HFI). NIC 832 may be embodied as or otherwise include any type of integrated circuit, discrete circuit, controller chip, chipset, add-in board, daughter card, network interface card, or other device that can be used by the computing sled 800 to connect to another computing device (e.g., with another sled 400). In some embodiments, NIC 832 may be embodied as part of a system-on-chip (SoC) that includes one or more processors, or included on a multi-chip package that also includes one or more processors. In some embodiments, NIC 832 may include a local processor (not shown) and / or local memory (not shown), both of which are local to NIC 832. In such embodiments, the local processor of NIC 832 may be capable of performing one or more of the functions of processor 820. Additionally or alternatively, in such embodiments, the local memory of the NIC 832 may be integrated into one or more components of the compute sled at the board level, socket level, chip level, and / or other levels.

[0056] The communication circuit 830 is communicatively coupled to an optical data connector 834. The optical data connector 834 is configured to mate with a corresponding optical data connector of the rack 240 when the compute sled 800 is installed in the rack 240. Illustratively, the optical data connector 834 includes a plurality of optical fibers leading from a mating surface of the optical data connector 834 to an optical transceiver 836. The optical transceiver 836 is configured to convert incoming optical signals from the rack-side optical data connector into electrical signals and convert electrical signals into outgoing optical signals to the rack-side optical data connector. While shown as forming part of the optical data connector 834 in the illustrative embodiment, in other embodiments, the optical transceiver 836 may form part of the communication circuit 830.

[0057] In some embodiments, the computing sled 800 may further include an expansion connector 840. In such embodiments, the expansion connector 840 is configured to mate with a corresponding connector of an extended chassisless circuit board substrate to provide additional physical resources to the computing sled 800. The additional physical resources may be used, for example, by the processor 820 during operation of the computing sled 800. The extended chassisless circuit board substrate may be substantially similar to the chassisless circuit board substrate 602 discussed above and may include various electrical components mounted thereto. The specific electrical components mounted to the extended chassisless circuit board substrate may depend on the intended functionality of the extended chassisless circuit board substrate. For example, the extended chassisless circuit board substrate may provide additional computing resources, memory resources, and / or storage resources. As such, the additional physical resources of the extended chassisless circuit board substrate may include, but are not limited to, processors, memory devices, storage devices, and / or accelerator circuits, including, for example, field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), security coprocessors, graphics processing units (GPUs), machine learning circuits, or other specialized processors, controllers, devices, and / or circuits.

[0058] Now refer to Figure 9 , shows an illustrative embodiment of a computing sled 800. As shown, a processor 820, communication circuitry 830, and optical data connectors 834 are mounted to the front face 650 of the chassisless circuit board substrate 602. The physical resources of the computing sled 800 can be mounted to the chassisless circuit board substrate 602 using any suitable attachment or mounting technique. For example, the various physical resources can be mounted in corresponding sockets (e.g., a processor socket), holders, or brackets. In some cases, some of the electrical components can be mounted directly to the chassisless circuit board substrate 602 via soldering or similar techniques.

[0059] As discussed above, separate processors 820 and communications circuitry 830 are mounted to the front face 650 of the chassisless circuit board substrate 602 so that the two heat-generating electrical components do not shadow each other. In the illustrative embodiment, the processor 820 and communications circuitry 830 are mounted in corresponding locations on the front face 650 of the chassisless circuit board substrate 602 so that two of those physical resources are not aligned in line with the other physical resources along the direction of the airflow path 608. It should be appreciated that while the optical data connector 834 is aligned with the communications circuitry 830, the optical data connector 834 does not generate heat or generates a nominal amount of heat during operation.

[0060] The memory devices 720 of the computing sled 800 are mounted to the reverse side 750 of the chassisless circuit board substrate 602, as discussed above with respect to the sled 400. Although mounted to the reverse side 750, the memory devices 720 are communicatively coupled to the processors 820 located on the reverse side 650 via the I / O subsystem 622. Because the chassisless circuit board substrate 602 is embodied as a double-sided circuit board, the memory devices 720 and the processors 820 can be communicatively coupled via one or more through-holes, connectors, or other mechanisms extending through the chassisless circuit board substrate 602. Of course, in some embodiments, each processor 820 can be communicatively coupled to a different set of one or more memory devices 720. Alternatively, in other embodiments, each processor 820 can be communicatively coupled to each memory device 720. In some embodiments, the memory devices 720 can be mounted to one or more memory mezzanines on the reverse side of the chassisless circuit board substrate 602 and can be interconnected with the corresponding processors 820 via a ball grid array.

[0061] Each of the processors 820 includes a heat sink 850 secured thereto. Due to the mounting of the memory devices 720 to the rear side 750 of the chassisless circuit board substrate 602 (and the vertical spacing of the corresponding sled 400 within the rack 240), the front side 650 of the chassisless circuit board substrate 602 includes additional "free" area or space, which facilitates the use of heat sinks 850 having a larger size relative to conventional heat sinks used in typical servers. Additionally, due to the improved thermal cooling characteristics of the chassisless circuit board substrate 602, none of the processor heat sinks 850 include a cooling fan attached thereto. In other words, each of the heat sinks 850 embodies a fanless heat sink.

[0062] Now refer to Figure 10In some embodiments, the sled 400 may be embodied as an accelerator sled 1000. The accelerator sled 1000 is optimized or otherwise configured to perform specialized computational tasks, such as machine learning, encryption, hashing, or other computationally intensive tasks. In some embodiments, for example, the compute sled 800 may offload tasks to the accelerator sled 1000 during operation. The accelerator sled 1000 includes various components similar to those of the sled 400 and / or the compute sled 800, which have been previously described. Figure 10 The same reference numbers are used to identify the above Figure 6 、 7 The descriptions of such components provided in and 8 are applicable to corresponding components of the accelerator skateboard 1000 and are not repeated herein for clarity of the description of the accelerator skateboard 1000.

[0063] In the illustrative accelerator sled 1000, the physical resource 620 is embodied as an accelerator circuit 1020. Figure 10 Only two accelerator circuits 1020 are shown, but it should be appreciated that in other embodiments the accelerator sled 1000 may include additional accelerator circuits 1020. For example, Figure 11 As shown in , in some embodiments, the accelerator sled 1000 may include four accelerator circuits 1020. The accelerator circuits 1020 may be embodied as any type of processor, coprocessor, computational circuit, or other device capable of performing computational or processing operations. For example, the accelerator circuits 1020 may be embodied as, for example, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a security coprocessor, a graphics processing unit (GPU), a machine learning circuit, or other specialized processor, controller, device, and / or circuit.

[0064] In some embodiments, the accelerator sled 1000 may also include an accelerator-to-accelerator interconnect 1042. Similar to the resource-to-resource interconnect 624 of the sled 600 discussed above, the accelerator-to-accelerator interconnect 1042 may be embodied as any type of communication interconnect capable of facilitating accelerator-to-accelerator communication. In an illustrative embodiment, the accelerator-to-accelerator interconnect 1042 is embodied as a high-speed point-to-point interconnect (e.g., faster than the I / O subsystem 622). For example, the accelerator-to-accelerator interconnect 1042 may be embodied as a Quick Path Interconnect (QPI), an Ultra Path Interconnect (UPI), or other high-speed point-to-point interconnect dedicated for processor-to-processor communication. In some embodiments, the accelerator circuit 1020 may be daisy-chained with a primary accelerator circuit 1020 connected to the NIC 832 and memory 720 via the I / O subsystem 622, and a secondary accelerator circuit 1020 connected to the NIC 832 and memory 720 via the primary accelerator circuit 1020.

[0065] Now refer to Figure 11 , shows an illustrative embodiment of an accelerator sled 1000. As discussed above, the accelerator circuit 1020, the communication circuit 830, and the optical data connector 834 are mounted to the front side 650 of the chassisless circuit board substrate 602. Again, the separate accelerator circuit 1020 and communication circuit 830 are mounted to the front side 650 of the chassisless circuit board substrate 602 so that the two heat-generating electrical components do not shadow each other, as discussed above. The memory device 720 of the accelerator sled 1000 is mounted to the rear side 750 of the chassisless circuit board substrate 602, as discussed above with respect to the sled 600. Although mounted to the rear side 750, the memory device 720 is communicatively coupled to the accelerator circuit 1020 located on the front side 650 via the I / O subsystem 622 (e.g., via vias). Furthermore, each of the accelerator circuits 1020 can include a heat sink 1070 that is larger than conventional heat sinks used in servers. As discussed above with reference to heat sink 870 , heat sink 1070 may be larger than conventional heat sinks due to the “free” area provided by memory devices 720 being located on the back side 750 of chassisless circuit board substrate 602 rather than on the front side 650 .

[0066] Now refer to Figure 12In some embodiments, the sled 400 may be embodied as a storage sled 1200. The storage sled 1200 is optimized or otherwise configured to store data in a data store 1250 that is local to the storage sled 1200. For example, during operation, the compute sled 800 or the accelerator sled 1000 may store and retrieve data from the data store 1250 of the storage sled 1200. The storage sled 1200 includes various components similar to those of the sled 400 and / or the compute sled 800, which have been previously described. Figure 12 The same reference numbers are used to identify the above Figure 6 、 7 The descriptions of such components provided in and 8 are applicable to the corresponding components of the storage sled 1200 and are not repeated herein for clarity of the description of the storage sled 1200.

[0067] In the illustrative storage sled 1200, the physical resource 620 is embodied as a storage controller 1220. Figure 12 Only two memory controllers 1220 are shown in the figure, but it should be appreciated that in other embodiments the memory sled 1200 may include additional memory controllers 1220. The memory controllers 1220 may be embodied as any type of processor, controller, or control circuit capable of controlling the storage and retrieval of data from the data storage 1250 based on requests received via the communication circuit 830. In the illustrative embodiment, the memory controllers 1220 are embodied as relatively low-power processors or controllers. For example, in some embodiments, the memory controllers 1220 may be configured to operate at a nominal power of approximately 75 watts.

[0068] In some embodiments, storage sled 1200 may also include a controller-to-controller interconnect 1242. Similar to the resource-to-resource interconnect 624 of sled 400 discussed above, controller-to-controller interconnect 1242 may be embodied as any type of communication interconnect capable of facilitating controller-to-controller communication. In an illustrative embodiment, controller-to-controller interconnect 1242 is embodied as a high-speed point-to-point interconnect (e.g., faster than I / O subsystem 622). For example, controller-to-controller interconnect 1242 may be embodied as a Quick Path Interconnect (QPI), an Ultra Path Interconnect (UPI), or other high-speed point-to-point interconnect specifically designed for processor-to-processor communication.

[0069] Now refer to Figure 13, shows an illustrative embodiment of a storage sled 1200. In the illustrative embodiment, data storage 1250 is embodied as or otherwise includes a storage cage 1252 configured to accommodate one or more solid-state drives (SSDs) 1254. To do so, storage cage 1252 includes a plurality of mounting slots 1256, each of which is configured to accommodate a corresponding solid-state drive 1254. Each of mounting slots 1256 includes a plurality of drive guides 1258 that cooperate to define an access opening 1260 for the corresponding mounting slot 1256. Storage cage 1252 is secured to chassisless circuit board substrate 602 such that the access opening faces away from chassisless circuit board substrate 602 (i.e., toward the front thereof). As such, when storage sled 1200 is installed in a corresponding rack 204, the solid-state drives 1254 are accessible. For example, the solid-state drive 1254 may be swapped out of the rack 240 (eg, via a robot) while the storage sled 1200 remains installed in the corresponding rack 240 .

[0070] The storage cage 1252 illustratively includes sixteen mounting slots 1256 and is capable of mounting and storing sixteen solid-state drives 1254. Of course, in other embodiments, the storage cage 1252 can be configured to store additional or fewer solid-state drives 1254. Additionally, in the illustrative embodiment, the solid-state drives are mounted vertically within the storage cage 1252, but in other embodiments, they can be mounted in a different orientation within the storage cage 1252. Each solid-state drive 1254 can embody any type of data storage device capable of storing long-term data. To do so, the solid-state drives 1254 can include the volatile and non-volatile memory devices discussed above.

[0071] like Figure 13 , the memory controller 1220, communication circuitry 830, and optical data connectors 834 are illustratively mounted to the front face 650 of the chassisless circuit board substrate 602. Again, as discussed above, the electrical components of the memory sled 1200 may be mounted to the chassisless circuit board substrate 602 using any suitable attachment or mounting technique, including, for example, sockets (e.g., a processor socket), supports, brackets, soldered connections, and / or other mounting or securing techniques.

[0072] As discussed above, the separate memory controller 1220 and communication circuitry 830 are mounted to the front face 650 of the chassisless circuit board substrate 602 so that the two heat-generating electrical components do not shadow each other. For example, the memory controller 1220 and communication circuitry 830 are mounted in corresponding positions on the front face 650 of the chassisless circuit board substrate 602 so that two of those electrical components are not aligned in line with the other along the direction of the airflow path 608.

[0073] The memory devices 720 of the storage sled 1200 are mounted to the rear side 750 of the chassisless circuit board substrate 602, as discussed above with respect to the sled 400. Although mounted to the rear side 750, the memory devices 720 are communicatively coupled to the storage controllers 1220 located on the front side 650 via the I / O subsystem 622. Again, because the chassisless circuit board substrate 602 embodies a double-sided circuit board, the memory devices 720 and the storage controllers 1220 can be communicatively coupled via one or more through-holes, connectors, or other mechanisms extending through the chassisless circuit board substrate 602. Each of the storage controllers 1220 includes a heat sink 1270 secured thereto. As discussed above, due to the improved thermal cooling characteristics of the chassisless circuit board substrate 602 of the storage sled 1200, none of the heat sinks 1270 includes a cooling fan attached thereto. In other words, each of the heat sinks 1270 embodies a fanless heat sink.

[0074] Now refer to Figure 14 In some embodiments, the sled 400 may be embodied as a memory sled 1400. The memory sled 1400 is optimized or otherwise configured to provide other sleds 400 (e.g., the compute sled 800, the accelerator sled 1000, etc.) with access to a pool of memory local to the memory sled 1200 (e.g., in two or more sets 1430, 1432 of the memory devices 720). For example, during operation, the compute sled 800 or the accelerator sled 1000 may remotely write to and / or read from one or more of the memory sets 1430, 1432 of the memory sled 1200 using a logical address space that maps to physical addresses in the memory sets 1430, 1432. The memory sled 1400 includes various components similar to those of the sled 400 and / or the compute sled 800, which have been described in detail in

[15] . Figure 14 The same reference numbers are used to identify the above Figure 6 、 7 The descriptions of such components provided in and 8 apply to the corresponding components of the memory sled 1400 and are not repeated herein for clarity of the description of the memory sled 1400.

[0075] In the illustrative memory sled 1400, the physical resource 620 is embodied as a memory controller 1420. Figure 14Only two memory controllers 1420 are shown in FIG. 1 , but it should be appreciated that in other embodiments the memory sled 1400 may include additional memory controllers 1420. The memory controllers 1420 may be embodied as any type of processor, controller, or control circuitry capable of controlling the writing and reading of data to and from the memory sets 1430, 1432 based on requests received via the communication circuitry 830. In the illustrative embodiment, each memory controller 1220 connects to a corresponding memory set 1430, 1432 to write to and read from the memory devices 720 within the corresponding memory set 1430, 1432, and implements any permissions (e.g., read, write, etc.) associated with a sled 400 that has sent a request to the memory sled 1400 to perform a memory access operation (e.g., read or write).

[0076] In some embodiments, memory sled 1400 may also include a controller-to-controller interconnect 1442. Similar to the resource-to-resource interconnect 624 of sled 400 discussed above, controller-to-controller interconnect 1442 may be embodied as any type of communication interconnect capable of facilitating controller-to-controller communication. In an illustrative embodiment, controller-to-controller interconnect 1442 is embodied as a high-speed point-to-point interconnect (e.g., faster than I / O subsystem 622). For example, controller-to-controller interconnect 1442 may be embodied as a Quick Path Interconnect (QPI), an Ultra Path Interconnect (UPI), or other high-speed point-to-point interconnect dedicated for processor-to-processor communication. As such, in some embodiments, a memory controller 1420 may access memory within a memory set 1432 associated with another memory controller 1420 via controller-to-controller interconnect 1442. In some embodiments, a scalable memory controller is comprised of multiple smaller memory controllers, referred to herein as "chiplets," on a memory sled (e.g., memory sled 1400). The chiplets may be interconnected (e.g., using EMIB (Embedded Multi-Die Interconnect Bridge)). The combined chiplet memory controller can scale up to a relatively large number of memory controllers and I / O ports (e.g., up to 16 memory channels). In some embodiments, memory controller 1420 can implement memory interleaving (e.g., mapping one memory address to memory set 1430, mapping the next memory address to memory set 1432, and mapping a third address to memory set 1430, etc.). Interleaving can be managed within memory controller 1420 or from a CPU socket (e.g., of a compute sled 800) across a network link to memory sets 1430, 1432, and can improve the latency associated with performing memory access operations compared to accessing adjacent memory addresses from the same memory device.

[0077] Furthermore, in some embodiments, the memory sled 1400 can be connected to one or more other sleds 400 (e.g., in the same rack 240 or an adjacent rack 240) via waveguides using waveguide connectors 1480. In the illustrated embodiment, the waveguides are 64 mm waveguides that provide 16 Rx (i.e., receive) channels and 16 Rt (i.e., transmit) channels. In the illustrated embodiment, each channel is 16 GHz or 32 GHz. In other embodiments, the frequencies can be different. Using waveguides can provide high-throughput access to the memory pool (i.e., memory clusters 1430, 1432) to another sled (e.g., a sled 400 in the same rack 240 as the memory sled 1400 or an adjacent rack 240) without placing an additional load on the optical data connectors 834.

[0078] Now refer to Figure 15A system for executing one or more workloads (e.g., applications) can be implemented in accordance with data center 100. In an illustrative embodiment, system 1510 includes an orchestrator server 1520, which can be embodied as a managed node including a computing device (e.g., compute sled 800) executing management software (e.g., a cloud operating environment such as OpenStack), which is communicatively coupled to a plurality of sleds 400 including a number of compute sleds 1530 (e.g., each similar to compute sled 800), memory sleds 1540 (e.g., each similar to memory sled 1400), accelerator sleds 1550 (e.g., each similar to memory sled 1000), and storage sleds 1560 (e.g., each similar to storage sled 1200). One or more of sleds 1530, 1540, 1550, 1560 may be grouped (such as by orchestrator server 1520) into managed nodes 1570 to collectively execute workloads (e.g., applications 1532 executed in virtual machines or containers). Managed nodes 1570 may be embodied as assemblies of physical resources 620, such as processors 820, memory resources 720, accelerator circuits 1020, or data storage 1250, from the same or different sleds 400. Furthermore, managed nodes may be established, defined, or "spun up" by orchestrator server 1520 when a workload is to be assigned to the managed node or at any other time, and may exist regardless of whether any workload is currently assigned to the managed node. In an illustrative embodiment, orchestrator server 1520 may selectively allocate and / or de-allocate physical resources 620 from sleds 400 and / or add or remove one or more sleds 400 from managed nodes 1570 based on quality of service (QoS) targets (e.g., performance targets associated with throughput, latency, instructions per second, etc.) associated with a service level agreement for a workload (e.g., application 1532). In doing so, orchestrator server 1520 may receive telemetry data indicating performance conditions (e.g., throughput, latency, instructions per second, etc.) in each sled 400 of managed nodes 1570 and compare the telemetry data to the quality of service targets to determine whether the quality of service targets are being met. If so, orchestrator server 1520 may additionally determine whether one or more physical resources can be de-allocated from managed node 1570 while still meeting the QoS targets, thereby freeing those physical resources for use in another managed node (e.g., to execute a different workload). Alternatively, if the QoS goals are not currently being met, orchestrator server 1520 can decide to dynamically allocate additional physical resources to help execute the workload (eg, application 1532 ) while the workload is executing.

[0079] Additionally, in some embodiments, orchestrator server 1520 can identify trends in resource utilization by workloads (e.g., application 1532), such as by identifying execution phases (e.g., time periods during which different operations are executed, each with different resource utilization characteristics) of the workload (e.g., application 1532) and preemptively identifying available resources in data center 100 and allocating them to managed nodes 1570 (e.g., within a predefined time period starting at the associated phase). In some embodiments, orchestrator server 1520 can model performance based on various latencies and distribution schemes to place workloads between compute sleds and other resources (e.g., accelerator sleds, memory sleds, storage sleds) in data center 100. For example, orchestrator server 1520 can utilize a model that accounts for the performance of resources on sled 400 (e.g., FPGA performance, memory access latency, etc.) as well as the performance of the path through the network to the resources (FPGA) (e.g., congestion, latency, bandwidth). As such, orchestrator server 1520 can determine which resource(s) should be used with respect to which workloads based on the total latency associated with each potential resource available in data center 100 (e.g., latency associated with the performance of the resource itself, in addition to latency associated with the path through the network between the compute sled executing the workload and sled 400 on which the resource is located).

[0080] In some embodiments, orchestrator server 1520 can use telemetry data reported from sled 400 (e.g., temperature, fan speed, etc.) to generate a heat generation map within data center 100 and allocate resources to managed nodes based on the heat generation map and the predicted heat generation associated with different workloads to maintain target temperatures and heat distribution within data center 100. Additionally or alternatively, in some embodiments, orchestrator server 1520 can organize the received telemetry data into a hierarchical model that indicates relationships between managed nodes (e.g., spatial relationships, such as the physical location of the resources of managed nodes within data center 100, and / or functional relationships, such as grouping managed nodes by customers served by them, the types of functions managed nodes typically perform, managed nodes that typically share or exchange workloads with each other, etc.). Based on differences in physical location and resources within managed nodes, a given workload can exhibit different resource utilization across the resources of different managed nodes (e.g., causing different internal temperatures, using different percentages of processor or memory capacity). Orchestrator server 1520 can determine the differences based on telemetry data stored in the hierarchical model and factor the differences into predictions of future resource utilization of the workload if the workload is reallocated from one managed node to another to accurately balance resource utilization in data center 100.

[0081] To reduce the computational load on orchestrator server 1520 and the data transfer load on the network, in some embodiments, orchestrator server 1520 can send self-test information to sleds 400 so that each sled 400 can locally (e.g., on sled 400) determine whether the telemetry data generated by sled 400 meets one or more conditions (e.g., available capacity meeting a predefined threshold, temperature meeting a predefined threshold, etc.). Each sled 400 can then report a simplified result (e.g., yes or no) back to orchestrator server 1520, which can utilize the simplified result when determining the allocation of resources to managed nodes.

[0082] Now refer to Figure 16 The system 1610 for dynamically adapting a communication protocol to network communications between endpoints may be implemented in accordance with the aforementioned reference. Figure 1 In an example embodiment, system 1610 includes an orchestrator server 1620 communicatively coupled to a plurality of sleds, including a compute sled 1630 and accelerator sleds 1640 , 1650 , and 1660 .

[0083] Compute sled 1630 and accelerator sleds 1640, 1650, and 1660, or portions thereof, may be grouped into managed nodes, such as by orchestrator server 1620. Managed nodes may collectively execute workloads, such as applications (e.g., application 1634). Managed nodes may be embodied as an assembly of resources (e.g., physical resources), such as compute resources, memory resources, storage resources, or other resources, from the same or different sleds or racks. As such, it should be appreciated that a sled may include multiple resources, each of which may be dedicated to a different managed node. Furthermore, managed nodes may be established, defined, or "spun up" by orchestrator server 1620 when a workload is to be assigned to a managed node or at any other time, and may exist regardless of whether any workload is currently assigned to the managed node. System 1610 may be located in a data center and provide storage and compute services (e.g., cloud services) to client devices 1614 communicating with system 1610 via network 1612. Orchestrator server 1620 may support a cloud operating environment, such as OpenStack, and managed nodes established by orchestrator server 1620 may execute one or more applications or processes (ie, workloads) on behalf of a user of client device 1614 , such as in a virtual machine or container.

[0084] Illustratively, compute sled 1630 includes one or more central processing units (CPUs) 1632 (e.g., processors or other devices or circuitry capable of performing a series of operations) that execute a workload (e.g., application 1634). Accelerator sled 1640 includes accelerator device 1642, accelerator sled 1650 includes accelerator device 1652, and accelerator sled 1660 includes accelerator device 1662. Each of accelerator devices 1642, 1652, or 1662 may be embodied as any device or circuitry that can be used to accelerate the execution of one or more operations. For example, the accelerator devices described herein may be embodied as any device or circuitry (e.g., a specialized processor, field programmable gate array (FPGA), application specific integrated circuit (ASIC), graphics processing unit (GPU), reconfigurable hardware, etc.) that can accelerate the execution of a portion of a workload, such as a workload task (e.g., a set of operations within a workload). Furthermore, each of the accelerator devices is configured with an accelerated kernel. Illustratively, accelerator device 1642 includes kernel 1644, accelerator device 1652 includes kernel 1654, and accelerator device 1662 includes kernel 1664. Each of the accelerated kernels may be embodied as a set of code or configuration of a portion of a corresponding accelerator device that causes the respective accelerator device to perform one or more accelerated functions (e.g., cryptographic operations, compression operations, etc.).

[0085] Each of the accelerator sleds 1640, 1650, and 1660 provides accelerated functions as a service to the workload processed by the managed node. In particular, each accelerator sled 1640, 1650, and 1660 can process requests from other sleds within the managed node (e.g., the compute sled 1630) to accelerate functions. For example, Figure 16 A compute sled 1630 is depicted executing an application 1634. Application 1634 may include functions to be executed sequentially. Compute sled 1630 may send a request to the accelerator sled to accelerate the execution of each function, thereby offloading the execution of the function to the accelerator device residing on the accelerator sled. The accelerator sled may respond to the request by providing a kernel on the accelerator device. For example, the accelerator sled may load a bitstream indicating the kernel into a slot (e.g., a subset of circuits or other logic units) on the accelerator device. Application 1634 may include a variety of functions that can be accelerated, such as cryptographic operations, machine learning algorithms, etc. The kernel provided on the accelerator device may be tailored to accelerate the execution of the corresponding function. For example, assume that the underlying function involves matrix multiplication. The kernel provided to the accelerator device may be specific to processing matrix multiplication operations. Once the kernel completes the acceleration of the function, it may return the resulting data to compute sled 1630. Orchestrator server 1620 may track (e.g., via a database) which kernels are registered with which accelerator sleds and accelerator devices.

[0086] In addition, system 1610 can expose a kernel-to-kernel communication network that allows any of kernels 1644, 1654, and 1664 to communicate with each other, for example, when sending processed workload data downstream to a kernel that processes a subsequent task in the workload. The kernel can establish a network connection via a given network communication protocol, such as TCP / IP (Transmission Control Protocol / Internet Protocol) or UDP (User Datagram Protocol). The kernel can encapsulate the workload data in one or more packets (datagrams in UDP) and transmit the packets (datagrams) to another kernel using the communication protocol.

[0087] Generally, network communication protocols can be characterized as either reliable or unreliable. Reliable protocols (e.g., TCP / IP) ensure that data transmitted by a sender reaches its intended recipient. Such protocols can notify the sender if a transmission fails (e.g., if a packet is dropped). However, reliable protocols typically incur overhead in determining whether a packet is successfully delivered and returning notification of delivery. Therefore, the operational cost of sending data via TCP / IP involves additional latency. In contrast, unreliable protocols (e.g., UDP) do not notify the sender if a transmission fails. However, because unreliable protocols generally lack the error checking and correction mechanisms otherwise provided by reliable protocols, they incur less overhead and are therefore more scalable than reliable protocols. Reliable protocols are often more desirable in scenarios where the probability of packet loss is relatively high, such as in scenarios where resource and network utilization in a system (e.g., system 1610) is high. Conversely, unreliable protocols can be used in scenarios where the probability of packet loss is relatively low, such as in scenarios where resource and network utilization is low.

[0088] As further described herein, embodiments of the present disclosure provide techniques for dynamically switching between reliable and unreliable protocols (and vice versa) for network communications (e.g., core-to-core communications) based on telemetry observed in system 1610. More specifically, an accelerator device in system 1610 (e.g., accelerator device 1642, 1652, or 1662) may include logic for receiving (or monitoring) telemetry data related, in part, to network utilization for core-to-core links. The telemetry data may include characteristics such as latency, throughput, and the current load on the underlying accelerator device(s) in inter-core communications. The accelerator device may evaluate the telemetry data against one or more conditions in a policy to determine whether to switch (e.g., change) the current network communication protocol to another. For example, suppose core A is currently transmitting data to core B using the UDP protocol. The accelerator device may observe that telemetry data indicating network utilization between cores A and B exceeds a certain threshold, triggering a condition in the policy. Because a reliable protocol may be more suitable for situations where network utilization is high, the policy may specify a change from UDP to a reliable protocol, such as TCP / IP.

[0089] Furthermore, over time, the accelerator device can learn patterns in telemetry data over time to predict instances of switching from one network communication protocol to another. For example, the accelerator device can use observed telemetry and time data as input to perform various machine learning techniques to generate prediction data. Based on subsequently observed telemetry data, the prediction data can indicate the likelihood that network communication for a given core link should switch from one core to another.

[0090] Now refer to Figure 17 The accelerator sled 1700 may be embodied as any type of computing device capable of performing the functions described herein, including monitoring telemetry data associated with network communications between accelerated cores, determining based on the monitored telemetry data that a condition is triggered to convert (e.g., change) the network communications from a given communication protocol to another communication protocol, and changing the network communications to the other communication protocol. The accelerator sled 1700 may represent a processor that is configured to execute a program that performs the operations described herein. Figure 16 Any of the accelerator slides 1640, 1650, or 1660 depicted in FIG.

[0091] like Figure 17 As shown in FIG, accelerator sled 1700 includes a computing engine 1702, an I / O subsystem 1708, a communication circuit 1710, one or more data storage devices 1714, and one or more accelerator devices 1718. Of course, in other embodiments, accelerator sled 1700 may include other or additional components, such as those typically found in computers (e.g., a display, peripheral devices, etc.). Additionally, in some embodiments, one or more of the illustrative components may be incorporated into or otherwise form part of another component.

[0092] Computing engine 1702 may be embodied as any type of device or collection of devices capable of performing the various computing functions described below. In some embodiments, computing engine 1702 may be embodied as a single device, such as an integrated circuit, an embedded system, an FPGA, a system on a chip (SoC), or other integrated system or device. Additionally, in some embodiments, computing engine 1702 includes or is embodied as a processor 1704 and a memory 1706. Processor 1704 may be embodied as any type of processor capable of performing the functions described herein. For example, processor 1704 may be embodied as a single-core or multi-core processor(s), a microcontroller, or other processor or processing / control circuitry. In some embodiments, processor 1704 may be embodied as, include, or be coupled to an FPGA, an ASIC, reconfigurable hardware or hardware circuitry, or other specialized hardware to facilitate the execution of the functions described herein.

[0093] Memory 1706 may be embodied as any type of volatile memory (e.g., dynamic random access memory (DRAM), etc.) or non-volatile memory or data storage capable of performing the functions described herein. Volatile memory may be a storage medium that requires power to maintain the state of the data stored by the medium. Non-limiting examples of volatile memory may include various types of random access memory (RAM), such as DRAM or static random access memory (SRAM). One specific type of DRAM that may be used in the memory module is synchronous dynamic random access memory (SDRAM). In certain embodiments, the DRAM of the memory component may comply with standards promulgated by JEDEC, such as JESD79F for DDR SDRAM, JESD79-2F for DDR2 SDRAM, JESD79-3F for DDR3 SDRAM, JESD79-4A for DDR4 SDRAM, JESD209 for Low Power DDR (LPDDR), JESD209-2 for LPDDR2, JESD209-3 for LPDDR3, and JESD209-4 for LPDDR4 (these standards are available at www.jedec.org). Such standards (and similar standards) may be referred to as DDR-based standards, and the communication interface of a memory device that implements such standards may be referred to as a DDR-based interface.

[0094] In one embodiment, the memory device is a block addressable memory device, such as those based on NAND or NOR technology. The memory device may also include future generation non-volatile devices, such as three-dimensional cross-point memory devices (e.g., Intel 3D Xpoint TM Memory) or other byte-addressable, write-at-place non-volatile memory devices. In one embodiment, the memory device may be or include a memory device using chalcogenide glass, multi-threshold level NAND flash memory, NOR flash memory, single or multi-level phase change memory (PCM), resistive memory, nanowire memory, ferroelectric transistor random access memory (FeTRAM), antiferroelectric memory, magnetoresistive random access memory (MRAM) memory that incorporates memristor technology, resistive memory including metal oxide substrate, oxygen vacancy substrate, and conductive bridge random access memory (CB-RAM), or spin transfer torque (STT)-MRAM, a device based on spintronic magnetic junction memory, a device based on magnetic tunnel junction (MTJ), a device based on DW (domain wall) and SOT (spin-orbit transfer), a memory device based on semiconductor thyristors, or a combination of any of the above memory devices, or other memory. The memory device may refer to the die itself and / or a packaged memory product.

[0095] In some embodiments, 3D crosspoint memory (e.g., Intel 3D Xpoint TM Memory 1706 may include a transistor-free, stackable cross-point architecture in which memory cells are located at the intersection of word lines and bit lines and are individually addressable and in which bit storage is based on changes in bulk resistance. In some embodiments, all or part of memory 1706 may be integrated into processor 1704. In operation, memory 1706 may store various software and data used during operation.

[0096] The compute engine 1702 is communicatively coupled to the other components of the accelerator sled 1700 via an I / O subsystem 1708, which may be embodied as circuitry and / or components used to facilitate input / output operations with the compute engine 1702 (e.g., with the processor 1704 and / or memory 1706) and the other components of the accelerator sled 1700. For example, the I / O subsystem 1708 may be embodied as or otherwise include a memory controller hub, an input / output control hub, an integrated sensor hub, a firmware device, communication links (e.g., point-to-point links, bus links, wires, cables, optical guides, printed circuit board traces, etc.), and / or other components and subsystems used to facilitate input / output operations. In some embodiments, the I / O subsystem 1708 may form part of a system on a chip (SoC) and be incorporated into the compute engine 1702 along with the processor 1704, memory 1706, and one or more of the other components of the accelerator sled 1700.

[0097] The communication circuitry 1710 may be embodied as any communication circuitry, device, or collection thereof capable of enabling communication between the accelerator sled 1700 and another computing device (e.g., the computing sled 1630, the accelerator sleds 1640, 1650, and 1660, etc.) via the network 1612. The communication circuitry 1710 may be configured to implement such communication using any one or more communication technologies (e.g., wired or wireless communication) and associated protocols (e.g., Ethernet, Bluetooth®, Wi-Fi®, WiMAX, etc.).

[0098] Illustrative communication circuitry 1710 includes a network interface controller (NIC) 1712, which may also be referred to as a host fabric interface (HFI). NIC 1712 may be embodied as one or more add-in boards, daughter cards, network interface cards, controller chips, chipsets, or other devices that can be used by accelerator sled 1700 to connect to another computing device (e.g., orchestrator server 1620, compute sled 1630, accelerator sleds 1640, 1650, and 1660, etc.). In some embodiments, NIC 1712 may be embodied as part of a system-on-chip (SoC) that includes one or more processors, or included on a multi-chip package that also includes one or more processors. In some embodiments, NIC 1712 may include a local processor (not shown) and / or local memory (not shown), both of which are local to NIC 1712. In such embodiments, the local processor of NIC 1712 may be capable of performing one or more of the functions of compute engine 1702 described herein. Additionally or alternatively, in such embodiments, the local memory of the NIC 1712 may be integrated into one or more components of the accelerator sled 1700 at the board level, socket level, chip level, and / or other levels.

[0099] The one or more illustrative data storage devices 1714 may be embodied as any type of device configured for short-term or long-term storage of data, such as, for example, memory devices and circuitry, memory cards, hard disk drives (HDDs), solid-state drives (SSDs), or other data storage devices. Each data storage device 1714 may include a system partition that stores data and firmware code for the data storage device 1714. Each data storage device 1714 may also include an operating system partition that stores data files and executable files for the operating system.

[0100] The accelerator device 1718 may represent Figure 16 The accelerator devices in system 1610 depicted in FIG1 , such as any combination of accelerator devices 1642 , 1652 , or 1662 , may form an accelerator subsystem that includes one or more buses or other interfaces between the accelerator devices in accelerator sled 1700 to enable the accelerator devices to share data. Furthermore, each accelerator device 1718 may send data to other accelerator devices in system 1610 via NIC 1712 based on the kernel configuration defined by orchestrator server 1620 . Each accelerator device 1718 may be embodied as any device or circuitry capable of accelerating the execution of a function (e.g., a specialized processor, FPGA, ASIC, GPU, reconfigurable hardware, etc.).

[0101] Additionally or alternatively, the accelerator sled 1700 may include one or more peripherals 1716. Such peripherals 1716 may include any type of peripherals typically found in computing devices, such as a display, speakers, mouse, keyboard, and / or other input / output devices, interface devices, and / or other peripherals.

[0102] Orchestrator server 1620, client device 1614, and compute sled 1630 may have Figure 17 The description of those components of accelerator sled 1700 applies equally to the description of the components of those devices and is not repeated herein for clarity of description. Further, it should be appreciated that client device 1614, orchestrator server 1620, and any of sleds 1630, 1640, 1650, and 1660 may include other components, subcomponents, and devices commonly found in computing devices that are not discussed above with reference to accelerator sled 1700 and are not discussed herein for clarity of description.

[0103] As described above, client device 1614, orchestrator server 1620, and sleds 1630, 1640, 1650, and 1660 illustratively communicate via network 1612, which can be embodied as any type of wired or wireless communication network, including a global network (e.g., the Internet), a local area network (LAN) or wide area network (WAN), a cellular network (e.g., Global System for Mobile Communications (GSM), 3G, Long Term Evolution (LTE), Worldwide Interoperability for Microwave Access (WiMAX)), etc.), a digital subscriber line (DSL) network, a cable network (e.g., a coaxial network, a fiber optic network, etc.), or any combination thereof.

[0104] Now refer to Figure 18During operation, accelerator sled 1700 can establish environment 1800. Of course, any of accelerator sleds 1640, 1650, and 1660 can similarly establish environment 1800 during operation. Illustratively, environment 1800 includes a network communicator 1820 and a protocol manager 1830. Each of the components of environment 1800 can be embodied as hardware, firmware, software, or a combination thereof. As such, in some embodiments, one or more of the components of environment 1800 can be embodied as a collection of circuits or electrical devices (e.g., network communicator circuit 1820, protocol manager circuit 1830, etc.). It should be appreciated that in such embodiments, one or more of network communicator circuit 1820 or protocol manager circuit 1830 can form part of one or more of compute engine 1702, communications circuit 1710, I / O subsystem 1708, accelerator device 1718, and / or other components of accelerator sled 1700. As shown, environment 1800 includes kernel configuration data 1802, which may be embodied as any data indicating a mapping of kernel configurations in system 1610 relative to workload flows. Kernel configuration data 1802 may also indicate the network communication protocol used by kernel-to-kernel links in system 1610. Furthermore, environment 1800 includes policy data 1804, which may be embodied as any data indicating a protocol change policy, including one or more change conditions evaluated based on monitored resource and network utilization. Furthermore, environment 1800 includes telemetry data 1806, which may be embodied as any data indicating the observed performance of accelerator sled 1700, accelerator device 1718, and other accelerator sleds and devices in system 1610 (e.g., power consumption, number of kernel connections, latency, average duration of connections, amount of data transferred per connection, etc.). Furthermore, environment 1800 includes prediction data 1808, which may be embodied as any data indicating the likelihood of triggering a protocol change policy condition based on subsequently observed telemetry data 1806.

[0105] The network communicator 1820, which, as discussed above, can be embodied as hardware, firmware, software, virtualized hardware, an emulated architecture, and / or a combination thereof, is configured to facilitate inbound and outbound network communications (e.g., network traffic, network packets, network flows, etc.) to and from the accelerator sled 1700, respectively. To do so, the network communicator 1820 is configured to receive and process data packets from one system or computing device (e.g., accelerator sled 1640, 1650, or 1660), and to prepare and send data packets to another computing device or system (e.g., compute sled 1630 or another accelerator sled 1640, 1650, or 1660). Accordingly, in some embodiments, at least a portion of the functionality of the network communicator 1820 can be performed by the communication circuitry 1710 and, in the illustrative embodiment, by the NIC 1712.

[0106] The protocol manager 1830, which can be embodied as hardware, firmware, software, virtualized hardware, an emulated architecture, and / or a combination thereof, is configured to monitor telemetry data 1806 associated with one or more network communications between a core and another core, wherein the network communications are established via a given communication protocol. The protocol manager 1830 is further configured to determine, based on the monitored telemetry data 1806, whether a condition in the policy data 1804 has been triggered to change the network communications from a current communication protocol to another communication protocol. The protocol manager 1830 is further configured to change the network communications between the cores to another communication protocol. As shown, the protocol manager 1830 includes a monitor component 1832, a selector component 1834, a configuration component 1836, and a predictor component 1838.

[0107] In some embodiments, monitor component 1832 is configured to obtain telemetry data 1806 associated with a network channel between a given core and another core. More specifically, a core configured within a given accelerator device 1818 can be interconnected with another core. The other core can be configured with the same accelerator device 1818 or another accelerator device in system 1610. Cores can be interconnected using a variety of schemes. For example, assume that another core is configured on an accelerator device on sled 1700. The cores can communicate with each other via NIC 1712. As another example, the other core may be configured on another accelerator sled in system 1610. In such a case, the cores can be interconnected via a switch device in system 1610 that interconnects the sled with sled 1300. Monitor component 1832 can obtain telemetry related to the link between the cores from NIC 1712 or the switch device in system 1610. For example, a resource monitor can reside in NIC 1712 or the switch device and obtain raw metrics regarding performance and utilization and send the metrics to monitor component 1832. In turn, the monitor component 1832 receives the raw metrics and can normalize the metrics to generate telemetry data 1806. Normalizing the metrics can involve converting the raw metrics into values ​​and types that can be further evaluated by the protocol manager 1830. In other embodiments, the monitor component 1832 is also configured to obtain telemetry data 1806 associated with the system 1610, such as the current load on the system 1610, the average network utilization between kernel connections, the average packet loss in kernel connections, etc.

[0108] In some embodiments, the selector component 1834 is configured to determine whether to convert (e.g., change) network communications between a core configured with respect to the accelerator device 1818 and another core to a different protocol based on an evaluation of the telemetry data 1806. For example, assume that network communications between core A and core B are currently performed via an unreliable protocol (such as UDP). The selector component 1834 can evaluate the telemetry data 1806 relative to the policy data 1804 to determine whether one or more conditions for changing to a reliable protocol (such as TCP / IP) are triggered. For example, the policy data 1804 can specify the following condition: if network utilization between the cores exceeds a specified threshold, then communications between the cores should be performed via a reliable protocol to ensure that core B receives data regardless of any additional latency resulting from the use of the reliable protocol.

[0109] In some embodiments, configuration component 1836 is configured to change the network communication between the cores to a different protocol if selector component 1834 determines that this is the case. Configuration component 1836 can modify kernel configuration data 1802 to indicate that a different protocol is to be used for the link between a given kernel on an accelerator device 1818 and another kernel, as determined by selector component 1834. Configuration component 1836 can also notify orchestrator server 1620 of the change in protocol. To do so, configuration component 1836 can send a message to orchestrator server 1620 identifying the kernel, the accelerator device on which the kernel is configured, and the protocol. In response, orchestrator server 1620 can propagate the update to other sleds and network devices in system 1610.

[0110] In some embodiments, predictor component 1838 is configured to learn one or more patterns based on telemetry data 1806 and changes between protocols over time to more quickly identify instances in which a change from a given protocol to another will be performed. For example, predictor component 1838 may execute various machine learning algorithms (e.g., optimization-based machine learning algorithms, prediction algorithms, etc.) and receive as input telemetry data 1806 related to a given core-to-core network link and system 1610. Predictor component 1838 may also receive timestamp inputs that define instances and periods in which a network link changed from one communication protocol to another. The machine learning algorithm may thus generate prediction data 1808. Selector component 1834 may further be configured to evaluate prediction data 1808 when determining whether to change to a different network communication protocol for a given core-to-core link. For example, selector component 1834 may retrieve subsequently collected telemetry data 1806 from monitor component 1832. The selector component 1834 can input the telemetry data 1806 to a machine learning algorithm, which can evaluate the telemetry data 1806 against the prediction data 1808. The results can indicate whether to change the currently configured network communication protocol to another network communication protocol.

[0111] Furthermore, the predictor component 1838 can include multiple prediction algorithms and provide a ranking of the algorithms for selection based on the execution of the telemetry data 1806 for each algorithm. For example, the ranking can be based on the percentage of each algorithm's results that converge toward the best possible result. Once a selection is provided, the predictor component 1838 can continue to use the selected algorithm in subsequent calculations.

[0112] Referring now to Figures 19A and 19B, diagrams of example embodiments of a core-to-core communication network are shown. In particular, Figures 19A and 19B illustrate a core-to-core communication network in which techniques for dynamically adapting reliable and unreliable network communication protocols can be implemented. Figure 19A depicts intercommunication between various cores in system 1610. In particular, Figure 19A includes accelerator sleds 1902 and 1912, which represent any of the accelerator sleds 1640, 1650, 1660, or 1700 in system 1610. Accelerator sled 1902 provides accelerator devices 1903 and 1907, and accelerator sled 1912 provides accelerator device 1913. Further, each of the accelerator devices includes one or more slots for loading one or more accelerated cores. For example, accelerator device 1903 includes slot 1904, accelerator device 1907 includes slot 1908, and accelerator device 1913 includes slot A 1914 and slot B 1917. Furthermore, each of the slots is configured with an accelerated core. Illustratively, slot 1904 is configured with core A 1905 and core B 1906; slot 1908 is configured with core C 1909 and core D 1910; slot A 1914 is configured with core E 1915 and core F 1916; and slot B is configured with core G 1918 and core H 1919.

[0113] As indicated by the double-headed arrows, each of the cores can communicate with each other via a core-to-core network exposed by system 1610. Specifically, system 1610 can provide a core network configuration that includes a NIC in each of accelerator sleds 1902 and 1912 and includes a network device (e.g., a network switch) that interconnects accelerator sleds 1902 and 1912. Orchestrator server 1620 can configure each of the NICs and network devices so that a given core sends data to another core based on workload flow. The NICs and network devices can form an orchestrator subsystem interface that connects components of the core's accelerator devices to each other to form a core-to-core network. The accelerator subsystem interface can expose a virtual address space that allows cores to identify and communicate with each other in the network.

[0114] Figure 19BThe diagram illustrates an abstraction of communication between core A 1905 and cores B 1906, C 1909, and H 1919. Core A 1905 can communicate via network 1920 (e.g., the accelerator subsystem described above). For example, cores B 1906, C 1909, and H 1919 may represent cores to which core A 1905 transmits processed data downstream, depending on the portion of the workload being executed. Illustratively, core A 1905 may interconnect with each of the cores via TCP / IP or UDP. The network communication protocol used for a given core-to-core link can be determined based on the resources and network utilization associated with the link. For example, core A 1905 and core B 1906 may be interconnected via UDP. Using UDP may indicate a network link with relatively low utilization. With such utilization, packets are less likely to be dropped, and therefore, packet loss detection mechanisms are less necessary. In contrast, core A is interconnected with cores C 1909 and H 1919 via TCP / IP. Using TCP / IP may indicate a network link with relatively high network utilization. Under such utilization, packets are more likely to be dropped, and therefore TCP / IP providing a packet loss detection mechanism may be more desirable.

[0115] Now refer to Figure 20 During operation, the accelerator sled 1700 may execute method 2000 to adapt a communication protocol to a network connection between endpoints (e.g., kernel endpoints). Of course, accelerator sleds 1640, 1650, and 1660 may also execute method 2000. As shown, method 2000 begins in block 2002, where the accelerator sled 1700 establishes network communication between a kernel configured thereon (e.g., in accelerator device 1718) and another kernel in system 1610 via a given network communication protocol. For example, the accelerator sled 1700 may do so by evaluating the kernel configuration (e.g., kernel configuration data 1802) to determine the currently designated network communication protocol to be used for the communication link between the kernel and the other kernel. Kernel configuration data 1802 may specify whether the current protocol is a reliable protocol (e.g., TCP / IP) or an unreliable protocol (e.g., UDP).

[0116] In block 2004, the accelerator sled 1700 monitors telemetry data associated with the established communication. For example, the accelerator sled 1700 may collect raw metrics from the NIC 1712 and other network components that interconnect the cores. Once collected, the accelerator sled 1700 may further process the metrics for evaluation. In block 2006, the accelerator sled 1700 determines whether a change condition has been triggered. To do so, the accelerator sled 1700 may evaluate the telemetry data against policy data for switching between a given network communication protocol and another. For example, a change condition specified in a policy may dictate a change from UDP to TCP / IP if network utilization observed in the telemetry data exceeds a predefined threshold. If the condition in the policy has not been triggered, the method 2000 returns to block 2004, and the accelerator sled 1700 continues monitoring the telemetry data.

[0117] Otherwise, if the change condition in the policy is triggered, then in block 2008, the accelerator sled 1700 changes the network communication protocol used for inter-core communications. Specifically, in block 2010, the accelerator sled 1700 evaluates the monitored telemetry data and the currently used protocol for the core-to-core link against the policy. In block 2012, the accelerator sled 1700 determines, based on the policy, whether to change to a reliable protocol (e.g., TCP / IP) or an unreliable protocol (UDP). For example, the policy may specify that if the currently used protocol is an unreliable protocol and the network bandwidth exceeds a specified threshold for a specified duration, the network communication protocol is to be changed to a reliable protocol. As another example, the policy may specify that if the currently used protocol is a reliable protocol and the average packet loss falls below a specified threshold for a specified duration, the network communication protocol is to be changed to an unreliable protocol.

[0118] In block 2014, accelerator sled 1700 modifies the configuration of the core-to-core link based on the determination. For example, accelerator sled 1700 may do so by accessing a locally stored configuration (e.g., core configuration data 1802) and modifying the configuration to indicate the protocol to be used for the core link. Furthermore, accelerator sled 1700 may notify orchestrator server 1620 of the change in the communication protocol to be used for the core link. As a result, orchestrator server 1620 may propagate the configuration change to other accelerator sleds in system 1610 to preserve integrity. In block 2016, accelerator sled 1700 establishes subsequent network communications between the cores using the protocol determined based on the policy.

[0119] Now refer to Figure 21During operation, the accelerator sled 1700 may apply machine learning techniques to determine whether to change between an unreliable protocol and a reliable protocol (or vice versa). In block 2018, the accelerator sled 1700 learns one or more change patterns based on the monitored telemetry data. For example, in block 2020, the accelerator sled 1700 evaluates the monitored telemetry data associated with the core-to-core link relative to a given point in time (e.g., as indicated by a timestamp) at which the accelerator sled 1700 changes from one communication protocol to another. In block 2022, the accelerator sled 1700 identifies patterns based on the evaluated telemetry data and time. For example, the accelerator sled 1700 may identify a tuple of telemetry values ​​at a previous point in time at which the accelerator sled 1700 changed to another network communication protocol. The accelerator sled 1700 may also identify additional points in time at which the tuple of telemetry values ​​triggered a change to another protocol. The identified points in time may indicate patterns associated with the change to another protocol.

[0120] In block 2024, the accelerator sled 1700 generates prediction data based on the one or more learned patterns. The prediction data indicates the likelihood of triggering a policy condition to change from one protocol to another based on subsequently observed telemetry data 1806. The prediction data can reduce the amount of telemetry data actually observed before changing to the other protocol, thereby improving network utilization. In block 2026, the accelerator sled 1700 uses the prediction data to determine a subsequent change from the network communication protocol to the other protocol. For example, the accelerator sled 1700 may return to the beginning of method 2000 and, in addition to evaluating subsequently monitored telemetry data against the policy, further evaluate the telemetry data against the prediction data. For example, the accelerator sled 1700 may observe a given tuple of telemetry data at a given point in time during the execution of a workload that would not otherwise trigger a change condition. However, after evaluating the tuple against the prediction, the accelerator sled 1700 may identify the tuple as the beginning of a pattern, thereby causing a change between protocols. Once identified, the accelerator sled 1700 may preemptively change protocols.

[0121] Example

[0122] The following provides illustrative examples of the technology disclosed herein. Embodiments of the technology may include any one or more, and any combination, of the examples described below.

[0123] Example 1 includes a skateboard comprising a computing engine that monitors telemetry data associated with one or more network communications between a first core of the skateboard and a second core configured on a second skateboard, wherein the one or more network communications are established via a first communication protocol, determines based on the monitored telemetry data that a condition for changing the network communication from the first communication protocol to the second communication protocol is triggered, and changes the network communication from the first communication protocol to the second communication protocol.

[0124] Example 2 includes the subject matter of Example 1, and wherein the network communications are to be changed from the first communications protocol to the second communications protocol comprises establishing subsequent network communications between the first kernel and the second kernel using the second communications protocol.

[0125] Example 3 includes the subject matter of any of Examples 1 and 2, and wherein the second communication protocol is determined according to a policy and the monitored telemetry data, wherein the policy defines a plurality of conditions for changing from a given network communication protocol to another network communication protocol.

[0126] Example 4 includes the subject matter of any of Examples 1-3, and wherein the second communication protocol corresponds to one of TCP / IP (Transmission Control Protocol / Internet Protocol) or UDP (User Datagram Protocol).

[0127] Example 5 includes the subject matter of any of Examples 1-4, and wherein the computing engine is further to learn one or more change patterns from the monitored telemetry data, wherein each change pattern defines the telemetry data observed over time.

[0128] Example 6 includes the subject matter of any of Examples 1-5, and wherein the computing engine is further to generate prediction data based on the learned one or more change patterns, wherein the prediction data indicates a likelihood that, based on subsequently monitored telemetry data, the network communication is to transition from the first communication protocol to the second communication protocol, or that the network communication is to transition from the second communication protocol to the first communication protocol.

[0129] Example 7 includes the subject matter of any of Examples 1-6, and wherein determining that the condition to change the network communication is triggered is further determined based on the prediction data.

[0130] Example 8 includes the subject matter of any of Examples 1-7, and wherein learning one or more change patterns from the monitored telemetry data comprises evaluating telemetry data associated with kernel network connections over time; and identifying the change patterns based on the evaluation.

[0131] Example 9 includes the subject matter of any of Examples 1-8, and wherein the computing engine is further to generate the prediction data via machine learning techniques.

[0132] Example 10 includes the subject matter of any of Examples 1-9, and wherein the computing engine is further to generate the prediction data via one or more machine learning techniques; and rank the prediction data according to each of the plurality of machine learning techniques.

[0133] Example 11 includes the subject matter of any of Examples 1-10, and wherein the telemetry data comprises at least one of a packet loss rate, a total amount of network connections, a throughput of the network connections, and a latency of the network connections.

[0134] Example 12 includes the subject matter of any of Examples 1-11, and wherein the compute engine is further to monitor telemetry data associated with one or more network connections between the first kernel and the third kernel, wherein the one or more network connections between the first kernel and the third kernel are established via a second communication protocol.

[0135] Example 13 includes the subject matter of any of Examples 1-12, and wherein the compute engine is further to change network communications between the first kernel and the third kernel to the first communication protocol based on the monitored telemetry data.

[0136] Example 14 includes the subject matter of any of Examples 1-13, and wherein the condition to change network communications is an indication that network utilization between the first core and the second core exceeds a specified threshold.

[0137] Example 15 includes a method comprising monitoring telemetry data associated with one or more network communications between a first core of a skateboard and a second core configured on a second skateboard, wherein the one or more network communications are established via a first communication protocol, determining based on the monitored telemetry data that a condition for changing the network communication from the first communication protocol to a second communication protocol is triggered, and changing the network communication from the first communication protocol to the second communication protocol.

[0138] Example 16 includes the subject matter of Example 15, and wherein changing the network communication from the first communication protocol to the second communication protocol comprises establishing subsequent network communication between the first kernel and the second kernel using the second communication protocol.

[0139] Example 17 includes the subject matter of any of Examples 15 and 16, and wherein the second communication protocol is determined according to a policy and the monitored telemetry data, wherein the policy defines a plurality of conditions for changing from a given network communication protocol to another network communication protocol.

[0140] Example 18 includes the subject matter of any of Examples 15-17, and wherein the second communication protocol corresponds to one of TCP / IP (Transmission Control Protocol / Internet Protocol) or UDP (User Datagram Protocol).

[0141] Example 19 includes the subject matter of any of Examples 15-18, and further includes learning one or more change patterns from the monitored telemetry data, wherein each change pattern defines observed telemetry data over time.

[0142] Example 20 includes the subject matter of any of Examples 15-19, and further includes generating predictive data based on the learned one or more change patterns, wherein the predictive data indicates a likelihood that, based on subsequently monitored telemetry data, network communications are to transition from a first communications protocol to a second communications protocol, or that network communications are to transition from a second communications protocol to the first communications protocol.

[0143] Example 21 includes the subject matter of any of Examples 15-20, and wherein determining that a condition to change network communications is triggered is further determined based on the prediction data.

[0144] Example 22 includes the subject matter of any of Examples 15-21, and wherein learning one or more change patterns from the monitored telemetry data comprises evaluating telemetry data associated with kernel network connections over time; and identifying patterns based on the evaluation.

[0145] Example 23 includes the subject matter of any of Examples 15-22, and further includes generating the prediction data via machine learning techniques.

[0146] Example 24 includes the subject matter of any of Examples 15-23, and further includes generating the prediction data via one or more machine learning techniques; and ranking the prediction data according to each of the plurality of machine learning techniques.

[0147] Example 25 includes the subject matter of any of Examples 15-24, and wherein the telemetry data comprises at least one of a packet loss rate, a total amount of network connections, a throughput of the network connections, and a latency of the network connections.

[0148] Example 26 includes the subject matter of any of Examples 15-25, and further includes monitoring telemetry data associated with one or more network connections between the first kernel and the third kernel, wherein the one or more network connections between the first kernel and the third kernel are established via a second communication protocol.

[0149] Example 27 includes the subject matter of any of Examples 15-26, and further includes changing network communications between the first core and the third core to a first communication protocol based on the monitored telemetry data.

[0150] Example 28 includes one or more machine-readable storage media including a plurality of instructions stored thereon that, in response to being executed, cause the skateboard to perform the method of any of Examples 15-27.

[0151] Example 29 includes a skateboard comprising components for performing the method of any of Examples 15-27.

[0152] Example 30 includes a skateboard comprising a computing engine to perform the method of any of Examples 15-17.

[0153] Example 31 includes a skateboard comprising a protocol manager circuit that monitors telemetry data associated with one or more network communications between a first core of the skateboard and a second core configured on a second skateboard, wherein the one or more network communications are established via a first communication protocol, determines based on the monitored telemetry data that a condition for changing the network communication from the first communication protocol to the second communication protocol is triggered, and changes the network communication from the first communication protocol to the second communication protocol.

[0154] Example 32 includes the subject matter of Example 31, and wherein the network communications are to be changed from the first communications protocol to the second communications protocol comprises establishing subsequent network communications between the first kernel and the second kernel using the second communications protocol.

[0155] Example 33 includes the subject matter of any of Examples 31 and 32, and wherein the second communication protocol is determined based on a policy and the monitored telemetry data, wherein the policy defines a plurality of conditions for changing from a given network communication protocol to another network communication protocol.

[0156] Example 34 includes the subject matter of any of Examples 31-33, and wherein the second communication protocol corresponds to one of TCP / IP (Transmission Control Protocol / Internet Protocol) or UDP (User Datagram Protocol).

[0157] Example 35 includes the subject matter of any of Examples 31-34, and wherein the protocol manager circuit is further to learn one or more change patterns from the monitored telemetry data, wherein each change pattern defines the telemetry data observed over time.

[0158] Example 36 includes the subject matter of any of Examples 31-35, and wherein the protocol manager circuit is further to generate prediction data based on the learned one or more change patterns, wherein the prediction data indicates a likelihood that, based on subsequently monitored telemetry data, the network communication is to transition from the first communication protocol to the second communication protocol, or that the network communication is to transition from the second communication protocol to the first communication protocol.

[0159] Example 37 includes the subject matter of any of Examples 31-36, and wherein determining that the condition to change the network communication is triggered is further determined based on the prediction data.

[0160] Example 38 includes the subject matter of any of Examples 31-37, and wherein learning one or more change patterns from the monitored telemetry data comprises evaluating telemetry data associated with kernel network connections over time; and identifying the change patterns based on the evaluation.

[0161] Example 39 includes the subject matter of any of Examples 31-38, and wherein the protocol manager circuit is further to generate the prediction data via machine learning techniques.

[0162] Example 40 includes the subject matter of any of Examples 31-39, and wherein the protocol manager circuit is further to generate the prediction data via one or more machine learning techniques; and rank the prediction data according to each of the plurality of machine learning techniques.

[0163] Example 41 includes the subject matter of any of Examples 31-40, and wherein the telemetry data comprises at least one of a packet loss rate, a total amount of network connections, a throughput of the network connections, and a latency of the network connections.

[0164] Example 42 includes the subject matter of any of Examples 31-41, and wherein the protocol manager circuit is further to monitor telemetry data associated with one or more network connections between the first core and the third core, wherein the one or more network connections between the first core and the third core are established via a second communication protocol.

[0165] Example 43 includes the subject matter of any of Examples 31-42, and wherein the protocol manager circuit is further to change network communications between the first core and the third core to the first communication protocol based on the monitored telemetry data.

[0166] Example 44 includes the subject matter of any of Examples 31-43, and wherein the condition to change network communications is an indication that network utilization between the first core and the second core exceeds a specified threshold.

[0167] Example 45 includes a skateboard comprising circuitry for monitoring telemetry data associated with one or more network communications between a first core of the skateboard and a second core configured on a second skateboard, wherein the one or more network communications are established via a first communication protocol, components for determining, based on the monitored telemetry data, that a condition is triggered to change the network communication from the first communication protocol to the second communication protocol, and components for changing the network communication from the first communication protocol to the second communication protocol.

[0168] Example 46 includes the subject matter of Example 45, and wherein the means for changing the network communication from the first communication protocol to the second communication protocol comprises circuitry for establishing subsequent network communications between the first core and the second core using the second communication protocol.

[0169] Example 47 includes the subject matter of any of Examples 45 and 46, and wherein the second communication protocol is determined according to a policy and the monitored telemetry data, wherein the policy defines a plurality of conditions for changing from a given network communication protocol to another network communication protocol.

[0170] Example 48 includes the subject matter of any of Examples 45-47, and wherein the second communication protocol corresponds to one of TCP / IP (Transmission Control Protocol / Internet Protocol) or UDP (User Datagram Protocol).

[0171] Example 49 includes the subject matter of any of Examples 45-48, and further includes means for learning one or more change patterns from the monitored telemetry data, wherein each change pattern defines observed telemetry data over time.

[0172] Example 50 includes the subject matter of any of Examples 45-49, and further includes a component for generating predictive data based on the learned one or more change patterns, wherein the predictive data indicates a likelihood that, based on subsequently monitored telemetry data, network communications are to transition from a first communications protocol to a second communications protocol, or that network communications are to transition from a second communications protocol to the first communications protocol.

[0173] Example 51 includes the subject matter of any of Examples 45-50, and wherein the means for determining that the condition to change the network communication is triggered is further determined based on the prediction data.

[0174] Example 52 includes the subject matter of any of Examples 45-51, and wherein the component for learning one or more change patterns from the monitored telemetry data includes circuitry for evaluating telemetry data associated with kernel network connections over time; and circuitry for identifying patterns based on the evaluation.

[0175] Example 53 includes the subject matter of any of Examples 45-52, and further includes means for generating the prediction data via machine learning techniques.

[0176] Example 54 includes the subject matter of any of Examples 45-53, and further includes means for generating the prediction data via one or more machine learning techniques; and means for ranking the prediction data according to each of the plurality of machine learning techniques.

[0177] Example 55 includes the subject matter of any of Examples 45-54, and wherein the telemetry data comprises at least one of a packet loss rate, a total amount of network connections, a throughput of the network connections, and a latency of the network connections.

[0178] Example 56 includes the subject matter of any of Examples 45-55, and further includes circuitry for monitoring telemetry data associated with one or more network connections between the first core and the third core, wherein the one or more network connections between the first core and the third core are established via a second communication protocol.

[0179] Example 57 includes the subject matter of any of Examples 45-56, and further includes means for changing network communications between the first core and the third core to the first communication protocol based on the monitored telemetry data.

Claims

1. A skateboard comprising: The calculation engine should: monitoring telemetry data associated with one or more network communications between a first core of the sled and a second core configured on a second sled, wherein the one or more network communications are established via a first communication protocol, wherein the first core and the second core are grouped into managed nodes by an orchestrator server based on telemetry data indicating performance conditions and quality of service objectives to jointly execute a workload, determining, based on the monitored telemetry data, that a condition is triggered to change network communications from a first communications protocol to a second communications protocol, and Changing network communications from a first communications protocol to a second communications protocol.

2. The skateboard of claim 1 , wherein changing the network communication from the first communication protocol to the second communication protocol comprises: Subsequent network communications are established between the first kernel and the second kernel using a second communication protocol.

3. The skateboard of claim 2, wherein the second communication protocol is determined based on a policy and the monitored telemetry data, wherein the policy defines a plurality of conditions for changing from a given network communication protocol to another network communication protocol.

4. The skateboard according to claim 1, wherein the second communication protocol corresponds to one of TCP / IP (Transmission Control Protocol / Internet Protocol) or UDP (User Datagram Protocol).

5. The skateboard of claim 1 , wherein the computing engine is further configured to: One or more change patterns are learned from the monitored telemetry data, where each change pattern defines observed telemetry data over time.

6. The skateboard of claim 5, wherein the computing engine further comprises: Predictive data is generated based on the learned one or more change patterns, wherein the predictive data indicates a likelihood that network communications are to transition from a first communications protocol to a second communications protocol, or vice versa, based on subsequently monitored telemetry data.

7. The skateboard of claim 6, wherein determining that a condition triggering a change in network communications is further determined based on predictive data.

8. The skateboard of claim 6 , wherein learning one or more change patterns from the monitored telemetry data comprises: Evaluating telemetry data associated with kernel network connections over time; and Based on the evaluation, patterns of change are identified.

9. The skateboard of claim 6, wherein the computing engine is further configured to: Generate predictive data through machine learning technology.

10. The skateboard of claim 6, wherein the computing engine is further configured to: generating predictive data via one or more machine learning techniques; and The predicted data is ranked according to each of the plurality of machine learning techniques.

11. The skateboard of claim 1 , wherein the telemetry data comprises at least one of a packet loss rate, a total amount of network connections, a throughput of network connections, and a latency of network connections.

12. The skateboard of claim 1 , wherein the computing engine is further configured to: Telemetry data associated with one or more network connections between the first core and the third core is monitored, wherein the one or more network connections between the first core and the third core are established via a second communication protocol.

13. The skateboard of claim 11 , wherein the computing engine is further configured to: Network communications between the first core and the third core are changed to a first communications protocol based on the monitored telemetry data.

14. The skateboard of claim 1, wherein the condition for changing network communications is an indication that network utilization between the first core and the second core exceeds a specified threshold.

15. A skateboard comprising: circuitry for monitoring telemetry data associated with one or more network communications between a first core of a sled and a second core configured on a second sled, wherein the one or more network communications are established via a first communication protocol, wherein the first core and the second core are grouped by an orchestrator server into managed nodes to jointly execute a workload based on telemetry data indicating performance conditions and quality of service objectives, means for determining, based on the monitored telemetry data, that a condition is triggered to change network communications from a first communications protocol to a second communications protocol, and Means for changing network communications from a first communications protocol to a second communications protocol.

16. The skateboard of claim 15, wherein the second communication protocol is determined based on a policy and the monitored telemetry data, wherein the policy defines a plurality of conditions for changing from a given network communication protocol to another network communication protocol.

17. The skateboard according to claim 15, further comprising: means for learning one or more change patterns from the monitored telemetry data, wherein each change pattern defines the telemetry data observed over time; as well as Components for generating predictive data based on the learned one or more change patterns, wherein the predictive data indicates a likelihood that network communications are to transition from a first communications protocol to a second communications protocol, or vice versa, based on subsequently monitored telemetry data.

18. The skateboard of claim 17, wherein the means for learning one or more changing patterns from the monitored telemetry data comprises: circuitry for evaluating telemetry data associated with kernel network connections over time; as well as A circuit for identifying a pattern based on the evaluation.

19. The skateboard according to claim 17, further comprising: means for generating predictive data via one or more machine learning techniques; as well as Means for ranking the prediction data according to each of the plurality of machine learning techniques.

20. The skateboard of claim 15, further comprising: means for monitoring telemetry data associated with one or more network connections between the first core and the third core, wherein the one or more network connections between the first core and the third core are established via a second communication protocol; as well as Means for changing network communications between the first core and the third core to a first communications protocol based on the monitored telemetry data.

21. A method comprising: monitoring telemetry data associated with one or more network communications between a first core of a sled and a second core configured on a second sled, wherein the one or more network communications are established via a first communication protocol, wherein the first core and the second core are grouped by an orchestrator server into managed nodes to jointly execute a workload based on telemetry data indicating a performance condition and a quality of service objective, determining, based on the monitored telemetry data, that a condition is triggered to change network communications from a first communications protocol to a second communications protocol, and Changing network communications from a first communications protocol to a second communications protocol.

22. The method of claim 21 , wherein changing network communications from a first communications protocol to a second communications protocol comprises: Subsequent network communications are established between the first kernel and the second kernel using a second communication protocol.

23. The method of claim 22, wherein the second communication protocol is determined based on a policy and the monitored telemetry data, wherein the policy defines a plurality of conditions for changing from a given network communication protocol to another network communication protocol.

24. A non-transitory machine-readable storage medium having instructions stored thereon that, when executed by a processor, cause the processor to perform the method according to any one of claims 21-23.

25. A computer program product having instructions which, when executed by a processor, cause the processor to perform the method according to any one of claims 21 to 23.

Images