A whole cabinet server
By introducing auxiliary management modules and chassis management controllers into the entire cabinet server, and using switching chips to optimize memory device allocation, the problem of scarce management interface resources is solved and the server's expansion capabilities and performance is improved.
Patent Information
- Application Number
- CN202411044948.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-31
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-07-31
AI Technical Summary
The chassis management controller management interface of the entire cabinet server is scarce, and it is impossible to effectively connect and manage multiple computing nodes, which limits the server's expansion capabilities and performance improvement.
By introducing auxiliary management modules and chassis management controllers into memory switching nodes, using management switching chips and network switching chips for data routing and resource allocation, generating offload or interconnect trigger instructions to optimize allocation and access of memory devices, reducing the need for The interface resource occupies the chassis management controller.
It improves the expansion capabilities and performance of the entire cabinet server, can better adapt to business growth or changes in demand, and enhances the management and configuration of computing nodes and storage resources.
Smart Images

Figure CN118567954B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of servers, and in particular to a whole cabinet server. Background Art
[0002] With the rapid development of technologies such as mobile, social media, cloud computing, and big data, the business that servers need to handle has exploded. In order to meet business needs, more servers are needed. The separate power supply and heat dissipation of each server will result in a very low energy efficiency ratio of the server. Choosing a whole cabinet server for power supply and heat dissipation can greatly improve the energy efficiency ratio, but the whole cabinet server system is complex. When monitoring it as a whole, it is usually necessary to interact with multiple management units. At present, the CMC (Chassis Management Controller) integrated in the chassis of the whole cabinet server is used to monitor and manage the hardware resources and power supply of the entire server chassis. However, the management resources of the CMC are limited. For a whole cabinet server with multiple computing nodes, the management interface resources of the CMC are scarce, resulting in the inability to connect to enough computing nodes. At the same time, the memory resource configuration and expansion capability of the whole cabinet server are limited. It is impossible to improve the overall performance or capacity by adding more computing nodes. When the business grows or the demand changes, the whole cabinet server may not be able to adapt.
[0003] Therefore, how to provide a solution to the above technical problems is a problem that those skilled in the art need to solve at present. Summary of the invention
[0004] The purpose of the present invention is to provide a whole cabinet server, which relates to the field of servers. In order to solve the problem of insufficient management interface resources of a chassis management controller, the whole cabinet server is supervised by jointly implementing an auxiliary management module and a chassis management controller, which can improve the performance and capacity of the whole cabinet server and the expansion capability of the whole cabinet server. When the business grows or the demand changes, the whole cabinet server can be better adapted.
[0005] To solve the above technical problems, the present invention provides a whole cabinet server, comprising a memory switch node and at least one computing node, wherein the computing node comprises a central processing unit and a baseboard management controller, the memory switch node comprises a switch module, a memory module and an auxiliary management module, the switch module comprises a chassis management controller, at least one management switch chip and at least one network switch chip, the chassis management controller and the baseboard management controller are connected to the network switch chip, and the memory module, the auxiliary management module and the central processing unit are connected to the management switch chip;
[0006] The auxiliary management module is used to determine the current allocation relationship between the plurality of computing nodes and the plurality of memory devices in the memory module according to the current routing configuration information of the management switch chip, and when detecting that there is a to-be-processed memory device that meets the preset processing condition, determine the target computing node corresponding to the to-be-processed memory device based on the current allocation relationship, and generate an unloading trigger instruction or an interconnection trigger instruction based on the preset processing condition, the target computing node and the to-be-processed memory device;
[0007] The chassis management controller is used to send an unloading execution instruction to the target computing node through the network switching chip when receiving the unloading trigger instruction, so that the target computing node performs a memory unloading operation; and when receiving the interconnection trigger instruction, control the first port and the second port of the management switching chip to be interconnected so that the target computing node accesses the memory device to be processed, the first port is a port connected to the memory device to be processed, and the second port is a port connected to the target computing node.
[0008] The preset processing condition includes a hot pull condition, the unloading trigger instruction includes a first unloading instruction and a second unloading instruction, and the process of generating an unloading trigger instruction or an interconnection trigger instruction based on the preset processing condition, the target computing node and the memory device to be processed includes:
[0009] When the preset processing condition satisfied by the to-be-processed memory device is the hot-plug condition, generating and sending the first unloading instruction based on the target computing node, and generating and sending the second unloading instruction based on the target computing node;
[0010] When receiving the unloading trigger instruction, sending the unloading execution instruction to the target computing node through the network switching chip so that the target computing node performs the memory unloading operation includes:
[0011] When receiving the second unloading instruction, sending an unloading execution instruction to the target computing node through the network switching chip, so that the target computing node performs a memory unloading operation;
[0012] The whole cabinet server further includes a power supply module, which is used to supply power to each of the memory devices, and cuts off the power supply to the memory device to be processed when receiving the first unloading instruction.
[0013] The power supply module includes a power supply and a plurality of electronic switches. The electronic switches are arranged on a power supply branch between the power supply and the memory device. The electronic switches are used to be disconnected when the first unloading instruction is received.
[0014] The auxiliary management module is further used to generate a prompt signal corresponding to executing a pull-out operation after the power supply of the memory device to be processed is cut off.
[0015] The preset processing condition includes an insertion condition, the interconnection trigger instruction includes a first interconnection instruction and a second interconnection instruction, and the process of generating an unloading trigger instruction or an interconnection trigger instruction based on the preset processing condition, the target computing node and the memory device to be processed includes:
[0016] When the preset processing condition satisfied by the memory device to be processed is the insertion condition, generating a first interconnection instruction according to the target computing node and the memory device to be processed, and generating a second interconnection instruction according to the memory device to be processed;
[0017] When the interconnection trigger instruction is received, the process of controlling the first port and the second port of the management switch chip to be interconnected so that the target computing node accesses the memory device to be processed includes:
[0018] Upon receiving the first interconnection instruction, controlling the first port and the second port of the management switch chip to be interconnected, so that the target computing node accesses the memory device to be processed;
[0019] The whole cabinet server further includes a power supply module, which is used to supply power to the memory device to be processed when the second interconnection instruction is received.
[0020] Among them, the power supply module includes a power supply and a plurality of electronic switches, wherein the electronic switch is arranged on a power supply branch between the power supply and the memory device, and the electronic switch is used to be turned on when receiving the second interconnection instruction so that the power supply supplies power to the memory device to be processed.
[0021] The auxiliary management module is further used to control the memory module to power on and output a first power-on instruction when receiving a power-on trigger instruction;
[0022] The chassis management controller is used to control each of the computing nodes to power on after receiving the first power-on instruction.
[0023] The memory exchange node further includes a first logic device, a first power supply and a plurality of electronic switches, and the computing node further includes a second logic device and a second power supply;
[0024] The first logic device is used to control the chassis management controller and the auxiliary management module to power on when receiving a power-on trigger instruction, and send the power-on trigger instruction to the auxiliary management module, control the management switch chip to power on when receiving a second power-on instruction, send a power-on instruction to the second logic device in the computing node corresponding to the third power-on instruction after receiving a third power-on instruction, and forward the power management signal to the chassis management controller after receiving a power management signal;
[0025] The auxiliary management module is further used to control the electronic switch to be turned on after receiving the power-on trigger instruction, and output a first power-on instruction at the same time;
[0026] The chassis management controller is further configured to output the second power-on instruction to the first logic device after receiving the first power-on instruction, and send the third power-on instruction to the first logic device after receiving the third power-on instruction;
[0027] The management switching chip is used to scan the central processing unit connected to itself after power-on, and generate the third power-on instruction corresponding to the central processing unit connected to itself;
[0028] The second logic device is used to control the startup of the computing node where it is located, and send the power management signal to the first logic device after the startup is completed.
[0029] Wherein, the first logic device is further used to control the memory module to perform a power-off protection operation when receiving a shutdown trigger instruction, and forward the shutdown trigger instruction to each of the second logic devices and the auxiliary management module;
[0030] The second logic device is further configured to execute a shutdown operation upon receiving the shutdown trigger instruction;
[0031] The auxiliary management module is also used to control the electronic switch to turn off when receiving the shutdown trigger instruction.
[0032] Wherein, the computing node further includes a fan;
[0033] The baseboard management controller is used to obtain and output the temperature data of the computing node where it is located;
[0034] The chassis management controller is also used to obtain temperature data of each of the computing nodes through the network switching chip, and control the rotation speed of the fan of each computing node according to a preset heat dissipation strategy.
[0035] The auxiliary management module is further used to configure the management switch chip to a memory mapping mode to obtain the type, connection rate and connection bandwidth of each port of the management switch chip.
[0036] The baseboard management controller is further used to set a static Internet Protocol address according to its own independent identification information, so that the chassis management controller can access the corresponding computing node according to the static Internet Protocol address.
[0037] Wherein, the memory device is a double-rate memory stick or a memory expansion card.
[0038] Wherein, the auxiliary management module is a microprocessor or a baseboard management controller migration module.
[0039] There are multiple network switching chips, including an internal network switching chip and an external network switching chip, the external network port of the chassis management controller is connected to the external network switching chip, the internal network port of the chassis management controller is connected to the internal network switching chip, the baseboard management controller is connected to the internal network switching chip, and the external management device is connected to the external network switching chip;
[0040] The external management device is used to output out-of-band management instructions;
[0041] The chassis management controller is further configured to manage the computing node according to the out-of-band management instruction.
[0042] The present invention provides a whole cabinet server, which realizes monitoring and management of the whole cabinet server through an auxiliary management module and a chassis management controller arranged on a memory exchange node. The memory exchange node is provided with a network exchange chip and a management exchange chip. The chassis management controller and multiple computing nodes are connected to the network exchange chip and the management exchange chip. The occupation of interface resources of the chassis management controller by the computing nodes is reduced through the exchange chip. The auxiliary management module and multiple memory devices in the memory module are connected to the management exchange chip, which is convenient for the auxiliary management module to allocate memory resources to each computing node. More computing nodes and storage resources can be added to improve the overall performance or capacity, which is convenient for improving the expansion capability of the whole cabinet server. When the business grows or the demand changes, the whole cabinet server can be better adapted. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] In order to more clearly illustrate the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0044] Figure 1 A schematic diagram of the structure of a whole cabinet server provided by the present invention;
[0045] Figure 2 A schematic diagram of identifying a port of a management switching chip provided by the present invention;
[0046] Figure 3 A schematic diagram of identifying a computing node or memory device for managing a switch chip provided by the present invention;
[0047] Figure 4 A schematic diagram of the structure of another whole cabinet server provided by the present invention;
[0048] Figure 5 A schematic diagram of network management of a whole cabinet server provided by the present invention;
[0049] Figure 6 This is a schematic diagram of network management of another whole cabinet server provided by the present invention. DETAILED DESCRIPTION
[0050] The core of the present invention is to provide a whole cabinet server, which relates to the server field. In order to solve the problem of insufficient management interface resources of the chassis management controller, the whole cabinet server is supervised by jointly implementing the auxiliary management module and the chassis management controller, which can improve the performance and capacity of the whole cabinet server and improve the expansion capability of the whole cabinet server. When the business grows or the demand changes, the whole cabinet server can be better adapted.
[0051] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0052] First, please refer to Figure 1 , Figure 1 The present invention provides a schematic diagram of the structure of a whole cabinet server, the whole cabinet server comprising:
[0053] A memory switching node and at least one computing node, the computing node includes a central processing unit 21 and a baseboard management controller 22, the memory switching node includes a switching module, a memory module and an auxiliary management module 12, the switching module includes a chassis management controller 111, at least one management switching chip 112 and at least one network switching chip 113, the chassis management controller 111 and the baseboard management controller 22 are connected to the network switching chip 113, and the memory module, the auxiliary management module 12 and the central processing unit 21 are connected to the management switching chip 112;
[0054] The auxiliary management module 12 is used to determine the current allocation relationship between the multiple computing nodes and the multiple memory devices 131 in the memory module according to the current routing configuration information of the management switch chip 112, and when it is detected that there is a to-be-processed memory device that meets the preset processing conditions, determine the target computing node corresponding to the to-be-processed memory device based on the current allocation relationship, and generate an unloading trigger instruction or an interconnection trigger instruction based on the preset processing conditions, the target computing node and the to-be-processed memory device;
[0055] The chassis management controller 111 is used to send an unloading execution instruction to the target computing node through the network switching chip 113 when receiving an unloading trigger instruction, so that the target computing node performs a memory unloading operation. When receiving an interconnection trigger instruction, the first port and the second port of the control management switching chip 112 are interconnected so that the target computing node accesses the memory device to be processed, the first port is a port connected to the memory device to be processed, and the second port is a port connected to the target computing node.
[0056] In this embodiment, all computing nodes are connected to the memory switch node. Specifically, the computing node mainly includes a motherboard, an OCP (Open Compute Project) module, an IO (Input Output) module, an E1.S (a server processor form) module, and a PCIe (Peripheral Component Interconnect Express) module. The motherboard is provided with a CPU (Central Processing Unit 21). Each CPU has 2 X16s connected to two CXL (Compute Express Link) Switch chips. Each computing node includes a BMC (Baseboard Management Controller) for implementing network configuration, power monitoring, and temperature monitoring of the computing node.
[0057] The memory exchange node mainly includes a SW (Switch) board, a management board and a memory module, wherein at least one exchange module is provided on the SW board, and the exchange module includes a chassis management controller 111 and at least one management exchange chip 112 and at least one network exchange chip 113. The management exchange chip 112 in this embodiment may be a CXL Switch chip. As an optional embodiment, two management exchange chips 112 may be provided on each SW board, and the CPUs of all computing nodes respectively provide 2 X16s to connect the two management exchange chips 112. The chassis management controller 111 supports power on and off of the entire machine, reset control, node identification, fan speed management, power management, out-of-band monitoring, and error integration. An auxiliary management module 12 is provided on the management board, which is arranged in the space above the SW board. The auxiliary management module 12 includes but is not limited to an MCPU (micro CPU, micro central processing unit, i.e., microprocessor) or a BMC migration module. The migration in this embodiment is used to represent detachability, i.e., the MCPU or BMC migration module is detachably arranged in the memory exchange node, and an upper component may be selected according to actual engineering needs, which is not limited in this embodiment.
[0058] It can be understood that the CPUs of all computing nodes are connected to the ports of the management switch chip 112 in the memory switch node, the BMCs of the computing nodes are connected to the ports of the network switch chip 113 in the memory switch node, and the memory module includes multiple memory devices 131, which are also connected to the ports of the management switch chip 112, so that the computing nodes can access the corresponding memory devices 131 through the management switch chip 112. Specifically, the management switch chip 112 implements data forwarding according to its routing configuration information, and the routing configuration information is used to limit the data received by a certain port of the management switch chip 112 to be output by another port of the management switch chip 112. When the CPU and the memory device 131 are connected to the port of the management switch chip 112, the allocation relationship between the CPU and the memory device 131 is determined. For example, it is assumed that the data received by the P1 port of the management switch chip 112 is forwarded to the P2 port according to the routing configuration information, that is, the memory device 131 connected to the P2 port is allocated to the CPU connected to the P1 port, and there is an allocation relationship between the two. In this embodiment, two ports with a routing forwarding relationship will be interconnected only when both are connected to a device (CPU or memory device 131). If any port is not connected to a device, the two ports are disconnected. The memory device 131 can be a DDR (Double Data Rate) or a memory expansion board, which can be selected according to actual engineering needs, and this embodiment does not limit it.
[0059] The auxiliary management module 12 can identify the first connection relationship between multiple computing nodes and each port of the management switching chip 112, and the second connection relationship between multiple memory devices 131 and each port of the management switching chip 112. According to the current routing configuration information, the first connection relationship and the second connection relationship of the management switching chip 112, the current allocation relationship between the memory device 131 and the central processing unit 21 can be determined. It can be understood that the current routing configuration information can be adjusted according to the actual project needs, thereby changing the routing status of each port of the management switching chip 112, and then adjusting the allocation relationship between the memory device 131 and the central processing unit 21, further adapting to the business demand changes, and allocating memory resources that meet the requirements to the computing nodes, thereby improving the overall performance and capacity. For example, under a certain routing configuration information, the CPU of the first computing node can access the first memory device 131, and under another routing configuration information, the CPU of the fifth computing node can access the first memory device 131.
[0060] When there is a memory device to be processed that meets the preset processing conditions in the memory devices 131 connected to the management switching chip 112, the target computing node corresponding to the memory device to be processed can be determined according to the current allocation relationship. In this embodiment, the preset processing conditions include hot-plug conditions and insertion conditions. The processing operations for the memory devices to be processed that meet different preset processing conditions are different. Therefore, according to the information of the memory device to be processed, the information of the target computing node and the preset processing conditions met by the memory device to be processed, corresponding operation trigger instructions are generated, including interconnection trigger instructions and unloading trigger instructions.
[0061] After receiving the unloading trigger instruction, the CMC triggers the BMC of the corresponding computing node through the network switching chip 113 to control the computing node to perform the unloading operation and unload the memory device to be processed. After receiving the interconnection trigger instruction, the CMC interconnects and controls the two corresponding ports of the management switching chip 112, so that the target computing node can access the memory device to be processed. The expansion of the memory device 131 and the computing node can be achieved through the CMC and the auxiliary management module 12, thereby improving the expansion capability of the cabinet server.
[0062] It can be seen that in this embodiment, the monitoring and management of the entire cabinet server are jointly realized by the auxiliary management module and the chassis management controller set on the memory switching node. The memory switching node is provided with a network switching chip and a management switching chip. The chassis management controller and multiple computing nodes are connected to the network switching chip and the management switching chip. The switching chip reduces the occupation of the interface resources of the chassis management controller by the computing nodes. The auxiliary management module and multiple memory devices in the memory module are connected to the management switching chip, which facilitates the auxiliary management module to allocate memory resources to each computing node. More computing nodes and storage resources can be added to improve the overall performance or capacity, which is convenient for improving the expansion capability of the entire cabinet server. When the business grows or the demand changes, the entire cabinet server can be better adapted.
[0063] Based on the above embodiments:
[0064] In an exemplary embodiment, the preset processing condition includes a hot pull condition, the unloading trigger instruction includes a first unloading instruction and a second unloading instruction, and the process of generating the unloading trigger instruction or the interconnection trigger instruction based on the preset processing condition, the target computing node, and the memory device to be processed includes:
[0065] When the preset processing condition satisfied by the memory device to be processed is a hot-plug condition, a first unloading instruction is generated based on the target computing node and sent, and a second unloading instruction is generated based on the target computing node and sent;
[0066] When receiving the unloading trigger instruction, the process of sending the unloading execution instruction to the target computing node through the network switching chip 113 so that the target computing node performs the memory unloading operation includes:
[0067] When receiving the second unloading instruction, sending an unloading execution instruction to the target computing node through the network switching chip 113, so that the target computing node performs the memory unloading operation;
[0068] The whole cabinet server also includes a power supply module, which is used to supply power to each memory device 131, and when receiving the first unloading instruction, cuts off the power supply of the memory device to be processed.
[0069] In this embodiment, the auxiliary management module 12 serves as a control module for hot-plugging the memory device 131, and controls the power on and off of the memory device 131. When the auxiliary management module 12 identifies that a certain memory device 131 fails or receives a memory removal instruction, it determines that the hot-plug condition is met, and generates a first unloading instruction to the CMC based on the target computing node. The CMC parses the information of the target computing node in the first unloading instruction, and sends an unloading execution instruction to the BMC of the target computing node to notify the target computing node to perform memory unloading. At the same time, the auxiliary management module 12 outputs a second unloading instruction to cut off the power supply of the memory device to be processed, and after cutting off the power supply of the memory device to be processed, generates a prompt signal corresponding to the execution of the unplugging operation, so that the user can unplug the memory device to be processed according to the prompt information.
[0070] In an exemplary embodiment, the power supply module includes a power supply and a plurality of electronic switches. The electronic switches are disposed on a power supply branch between the power supply and the memory device 131 . The electronic switches are configured to be disconnected upon receiving a first unloading instruction.
[0071] In this embodiment, an electronic switch is provided on the power supply branch between the power supply and the memory device 131, and the power on and off of the memory device 131 is controlled by controlling the on and off of the electronic switch. The electronic switch is disconnected under the control of the first unloading instruction, thereby cutting off the power supply of the power supply to the memory device 131. The electronic switch can select an electronic fuse efuse, which is resettable and does not need to be replaced, reducing maintenance costs and downtime, and has a fast response speed. At the same time, it can also implement overcurrent protection, further improving the operation safety of the entire cabinet server.
[0072] In an exemplary embodiment, the preset processing condition includes an insertion condition, the interconnection trigger instruction includes a first interconnection instruction and a second interconnection instruction, and the process of generating an unloading trigger instruction or an interconnection trigger instruction based on the preset processing condition, the target computing node, and the memory device to be processed includes:
[0073] When the preset processing condition satisfied by the memory device to be processed is an insertion condition, a first interconnection instruction is generated according to the target computing node and the memory device to be processed, and a second interconnection instruction is generated according to the memory device to be processed;
[0074] When receiving the interconnection trigger instruction, the process of controlling the first port and the second port of the management switch chip 112 to be interconnected so that the target computing node accesses the memory device to be processed includes:
[0075] When receiving the first interconnection instruction, controlling the first port and the second port of the management switch chip 112 to be interconnected, so that the target computing node accesses the memory device to be processed;
[0076] The whole cabinet server also includes a power supply module, which is used to supply power to the memory device to be processed when receiving the second interconnection instruction;
[0077] The power supply module includes a power supply and multiple electronic switches. The electronic switches are arranged on the power supply branch between the power supply and the memory device 131. The electronic switches are used to be turned on when receiving the second interconnection instruction so that the power supply can supply power to the memory device to be processed.
[0078] In this embodiment, when a new memory device 131 is inserted, the auxiliary management module 12 can detect the presence signal of the new memory device 131, and the auxiliary management module 12 sends an I2C instruction to control the electronic switch to turn on the power. After the new memory device 131 is initialized, the auxiliary management module 12 reads the memory resource information of the new memory device 131 and triggers the CMC so that the CMC controls the interconnection of the corresponding first port and the second port on the management switching chip 112.
[0079] In this embodiment, both the auxiliary management module 12 and the CMC can detect the specific configuration of the management switch chip 112 of the memory node, that is, which computing node is connected to which port of which management switch chip 112. For specific implementation, refer to Figure 2 As shown, the central processor 21 is connected to the input / output switching chip through a level conversion circuit, and the input / output switching chip is connected to the management switching chip 112 through two connectors. ID coding is performed on each management switching chip 112 and each port on the SW board. Figure 2 In the figure, S1 and S2 are used to indicate the codes of the management switch chip 112, and P0, P1, P2, and P3 are used to indicate the codes of the ports. Different IDs are set for each port. The central processor 21 of the computing node is configured as follows: Figure 2 The I2C (Inter-Integrated Circuit) link shown in the figure knows which port of the management switch chip 112 is connected. On the other hand, a pull-down resistor is connected to each port PRSNT of the management switch chip 112. When the port is in place, the auxiliary management module 12 Figure 3 The I2C link shown learns which computing node or which memory device 131 is connected through which port. It can be understood that the PRSNT signal indicates whether the port has been plugged into the corresponding device or cable. Figure 3 In the figure, N0-N7 represent the first computing node to the eighth computing node respectively, and PE4 and PE5 are used to represent the relevant characteristic information of the PCIe channel.
[0080] In an exemplary embodiment, referring to Figure 4 As shown, the auxiliary management module 12 is also used to control the memory module to power on and output a first power-on instruction when receiving a power-on trigger instruction;
[0081] The chassis management controller 111 is configured to control each computing node to power on upon receiving a first power-on instruction.
[0082] In an exemplary embodiment, the memory switch node further includes a first logic device, a first power supply, and a plurality of electronic switches, and the computing node further includes a second logic device and a second power supply;
[0083] The first logic device is used to control the chassis management controller 111 and the auxiliary management module 12 to power on when receiving a power-on trigger instruction, and send the power-on trigger instruction to the auxiliary management module 12, control the management switch chip 112 to power on when receiving a second power-on instruction, and after receiving a third power-on instruction, send the power-on instruction to the second logic device in the computing node corresponding to the third power-on instruction, and forward the power management signal to the chassis management controller 111 after receiving the power management signal;
[0084] The auxiliary management module 12 is also used to control the electronic switch to conduct after receiving the power-on trigger instruction, and output the first power-on instruction at the same time;
[0085] The chassis management controller 111 is further configured to output a second power-on instruction to the first logic device after receiving the first power-on instruction, and to send the third power-on instruction to the first logic device after receiving the third power-on instruction;
[0086] The management switching chip 112 is used to scan the central processing unit 21 connected to itself after power-on, and generate a third power-on instruction corresponding to the central processing unit 21 connected to itself;
[0087] The second logic device is used to control the startup of the computing node where it is located, and send a power management signal to the first logic device after the startup is completed.
[0088] In this embodiment, the CMC implements the power-on and power-off timing control of the memory exchange node and each computing node. It can be understood that after the PSU (Power Supply Unit, i.e., the power supply in this embodiment) of each node (memory exchange node and computing node) is powered, the VR (Voltage Regulator, voltage conversion chip) transfers each STBY power (Standby Voltage). When the STBY power of each node is completed, the first CPLD (Complex Programmable Logic Device, complex programmable logic device) (first logic device), BMC, second CPLD (second logic device) and CMC work normally, and the CPLD on the computing node and the memory exchange node sends the PG signal of the STBY power to its own BMC or CMC through I2C. After short pressing the power button of the memory exchange node, the power on trigger instruction is passed to the first CPLD, and the first CPLD controls the power on of other devices except the management exchange chip 112 and the memory module, and forwards the power on trigger instruction to the auxiliary management module 12. After the auxiliary management module 12 is turned on and enters the system, it automatically executes instructions: it controls the electronic switch through I2C to turn on the power of the memory device 131, and at the same time sends the LPC instruction to turn on the management switch chip 112 to the CMC, and the CMC forwards it to the first CPLD through I2C to turn on the power of the management switch chip 112. After the management switch chip 112 is turned on, the internal DSP will automatically scan the downstream device to learn the computing node connected to itself, and send the third power-on instruction to turn on the computing node to the CMC. The CMC forwards the third power-on instruction of the computing node to the first CPLD through I2C, and the first CPLD sends a PWREN signal to the second CPLD of each computing node corresponding to the third power-on instruction to control the power-on of each computing node. The second CPLD of the computing node sends PWRGD (power management signal) to the first CPLD of the memory switch node, and the first CPLD of the memory switch node then forwards PWRGD to the CMC through I2C.
[0089] In an exemplary embodiment, the first logic device is further used to control the memory module to perform a power-off protection operation upon receiving a shutdown trigger instruction, and forward the shutdown trigger instruction to each second logic device and the auxiliary management module 12;
[0090] The second logic device is also used to execute a shutdown operation when receiving a shutdown trigger instruction;
[0091] The auxiliary management module 12 is further used to control the electronic switch to turn off when receiving a shutdown trigger instruction.
[0092] In this embodiment, after long pressing the power button of the memory switch node, the power off command (shutdown trigger command) is transmitted to the first CPLD, controlling all VRs of the memory switch node to shut down the main power (main voltage). At the same time, the first CPLD sends a power off command to each computing node and the auxiliary management module 12 to perform a shutdown operation.
[0093] In an exemplary embodiment, the computing node further includes a fan;
[0094] The baseboard management controller 22 is used to obtain and output the temperature data of the computing node where it is located;
[0095] The chassis management controller 111 is also used to obtain temperature data of each computing node through the network switching chip 113, and control the rotation speed of the fan of each computing node according to a preset heat dissipation strategy.
[0096] CMC monitors the temperature sensors of the entire machine and controls the fan speed according to the cooling strategy to achieve cooling management of the entire machine. Of course, in addition to monitoring the temperature of each board and chip in the memory conversion node, CMC also needs to monitor the temperature sensors of each computing node BMC and the system health status sensor to achieve cooling management of the computing node.
[0097] In an exemplary embodiment, the auxiliary management module 12 is further used to configure the management switch chip 112 to a memory mapping mode to obtain the type, connection rate, and connection bandwidth of each port of the management switch chip 112 .
[0098] In this embodiment, the memory mapping mode of the CXL Switch is configured through the I2C of the MCPU / BMC migration module to obtain the type, link rate, and link bandwidth information of each port; the two groups of PCIe Gen3 x2 of the MCPU are respectively connected to the two Switch chips as in-band management ports to obtain CXL memory resource information and memory fault information and manage the memory mapping mode; the one group of PCIe Gen2 x1 of the BMC migration module is divided into two groups of PCIe Gen2 x1 through the PCIe SW and respectively connected to the two Switch chips as in-band management ports.
[0099] In an exemplary embodiment, there are multiple network switch chips 113, and the multiple network switch chips 113 include an internal network switch chip and an external network switch chip. The external network port of the chassis management controller 111 is connected to the external network switch chip, the internal network port of the chassis management controller 111 is connected to the internal network switch chip, the baseboard management controller 22 is connected to the internal network switch chip, and the external management device is connected to the external network switch chip.
[0100] An external management device for outputting out-of-band management instructions;
[0101] The chassis management controller 111 is also used to manage the computing nodes according to out-of-band management instructions.
[0102] Reference Figure 5 As shown, the management of multiple computing nodes within the CMC depends on network communication. To solve the problem that the management of multiple nodes in the cabinet depends on too many network ports, this embodiment separates the internal and external management networks, and manages the BMCs of multiple computing nodes by port forwarding by the CMC. To support the internal management of the entire chassis and the external management of the entire cabinet by Bracelet, the internal and external management networks are distinguished. The internal management network is responsible for the management of chassis nodes, and the external management network includes the out-of-band management network of the CMC and the independent out-of-band management network of each computing node.
[0103] In an exemplary embodiment, the baseboard management controller 22 is further configured to set a static Internet Protocol address according to its own independent identification information, so that the chassis management controller 111 can access the corresponding computing node according to the static Internet Protocol address.
[0104] Reference Figure 6 As shown, Figure 6 N0-N7 represent the first computing node to the eighth computing node respectively. Each computing node includes a baseboard management controller and a network card. The baseboard management controller is connected to the network card and is connected to the front-end Ethernet interface and the network switching chip 113 through the physical layer. The CMC module outputs 2 sets of serdes (Serializer / Deserializer) signals. The BMC of each computing node automatically sets a static IP (Internet Protocol) according to its own ID (Identification, independent identification information). The two MACs (Media Access Control) of the CMC are connected to the DHCP (Dynamic Host Configuration Protocol) network switch and the static network switch respectively. The CMC accesses the computing nodes and can access the static network through the dynamic network. Two RJ45 (a network interface connector) network ports are reserved on the entire cabinet, namely Figure 6The whole cabinet system management Ethernet interface and the whole cabinet chassis management controller management Ethernet interface support the MCPU system and CMC management network, which is convenient for the whole cabinet network management. The I / O board of the memory exchange node reserves an RJ45 network port, namely the system management Ethernet interface and the chassis management controller management Ethernet interface. The chassis management controller 111 includes a baseboard management controller, which is connected to two network switching chips 113 through two physical layers, supports the MCPU system and CMC management network, and is convenient for the management and allocation of cxl memory.
[0105] In summary, the present invention, on the basis of realizing the whole cabinet management function, makes maximum use of management interface resources, and combines CMC and MCPU to realize network management, whole machine power on and off management, node ID identification, temperature monitoring, and hot plug management functions.
[0106] It should also be noted that, in this specification, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "comprising" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.
[0107] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A whole cabinet server, characterized in that: It includes a memory switching node and at least one computing node, the computing node includes a central processing unit and a baseboard management controller, the memory switching node includes a switching module, a memory module and an auxiliary management module, the switching module includes a chassis management controller, at least one management switching chip and at least one network switching chip, the chassis management controller and the baseboard management controller are connected to the network switching chip, and the memory module, the auxiliary management module and the central processing unit are connected to the management switching chip; The auxiliary management module is used to determine the current allocation relationship between the plurality of computing nodes and the plurality of memory devices in the memory module according to the current routing configuration information of the management switch chip, and when detecting that there is a to-be-processed memory device that meets the preset processing condition, determine the target computing node corresponding to the to-be-processed memory device based on the current allocation relationship, and generate an unloading trigger instruction or an interconnection trigger instruction based on the preset processing condition, the target computing node and the to-be-processed memory device; The chassis management controller is configured to, upon receiving the unloading trigger instruction, send an unloading execution instruction to the target computing node through the network switching chip so that the target computing node performs a memory unloading operation, and upon receiving the interconnection trigger instruction, control the first port and the second port of the management switching chip to be interconnected so that the target computing node accesses the memory device to be processed, wherein the first port is a port connected to the memory device to be processed, and the second port is a port connected to the target computing node; The preset processing condition includes a hot pull condition, the unloading trigger instruction includes a first unloading instruction and a second unloading instruction, and the process of generating an unloading trigger instruction or an interconnection trigger instruction based on the preset processing condition, the target computing node and the memory device to be processed includes: When the preset processing condition satisfied by the to-be-processed memory device is the hot-plug condition, generating and sending the first unloading instruction based on the target computing node, and generating and sending the second unloading instruction based on the target computing node; When receiving the unloading trigger instruction, sending the unloading execution instruction to the target computing node through the network switching chip so that the target computing node performs the memory unloading operation includes: When receiving the second unloading instruction, sending an unloading execution instruction to the target computing node through the network switching chip, so that the target computing node performs a memory unloading operation; The whole cabinet server further includes a power supply module, which is used to supply power to each of the memory devices, and cuts off the power supply to the memory device to be processed when receiving the first unloading instruction.
2. The whole cabinet server according to claim 1, characterized in that: The power supply module includes a power supply and a plurality of electronic switches. The electronic switches are arranged on a power supply branch between the power supply and the memory device. The electronic switches are used to be disconnected when the first unloading instruction is received.
3. The whole cabinet server according to claim 2, characterized in that: The auxiliary management module is also used to generate a prompt signal corresponding to executing a pull-out operation after the power supply of the memory device to be processed is cut off.
4. The whole cabinet server according to claim 1, characterized in that: The preset processing condition includes an insertion condition, the interconnection trigger instruction includes a first interconnection instruction and a second interconnection instruction, and the process of generating an unloading trigger instruction or an interconnection trigger instruction based on the preset processing condition, the target computing node and the memory device to be processed includes: When the preset processing condition satisfied by the memory device to be processed is the insertion condition, generating a first interconnection instruction according to the target computing node and the memory device to be processed, and generating a second interconnection instruction according to the memory device to be processed; When the interconnection trigger instruction is received, the process of controlling the first port and the second port of the management switch chip to be interconnected so that the target computing node accesses the memory device to be processed includes: Upon receiving the first interconnection instruction, controlling the first port and the second port of the management switch chip to be interconnected, so that the target computing node accesses the memory device to be processed; The whole cabinet server further includes a power supply module, which is used to supply power to the memory device to be processed when the second interconnection instruction is received.
5. The whole cabinet server according to claim 4, characterized in that: The power supply module includes a power supply and a plurality of electronic switches. The electronic switch is arranged on a power supply branch between the power supply and the memory device. The electronic switch is used to be turned on when receiving the second interconnection instruction so that the power supply supplies power to the memory device to be processed.
6. The whole cabinet server according to claim 1, characterized in that: The auxiliary management module is further used to control the memory module to power on and output a first power-on instruction when receiving a power-on trigger instruction; The chassis management controller is used to control each of the computing nodes to power on after receiving the first power-on instruction.
7. The whole cabinet server according to claim 6, characterized in that: The memory exchange node also includes a first logic device, a first power supply and a plurality of electronic switches, and the computing node also includes a second logic device and a second power supply; The first logic device is used to control the chassis management controller and the auxiliary management module to power on when receiving a power-on trigger instruction, and send the power-on trigger instruction to the auxiliary management module, control the management switch chip to power on when receiving a second power-on instruction, send a power-on instruction to the second logic device in the computing node corresponding to the third power-on instruction after receiving a third power-on instruction, and forward the power management signal to the chassis management controller after receiving a power management signal; The auxiliary management module is further used to control the electronic switch to be turned on after receiving the power-on trigger instruction, and output a first power-on instruction at the same time; The chassis management controller is further configured to output the second power-on instruction to the first logic device after receiving the first power-on instruction, and send the third power-on instruction to the first logic device after receiving the third power-on instruction; The management switching chip is used to scan the central processing unit connected to itself after power-on, and generate the third power-on instruction corresponding to the central processing unit connected to itself; The second logic device is used to control the startup of the computing node where it is located, and send the power management signal to the first logic device after the startup is completed.
8. The whole cabinet server according to claim 7, characterized in that: The first logic device is further used to control the memory module to perform a power-off protection operation when receiving a shutdown trigger instruction, and forward the shutdown trigger instruction to each of the second logic devices and the auxiliary management module; The second logic device is further configured to execute a shutdown operation upon receiving the shutdown trigger instruction; The auxiliary management module is also used to control the electronic switch to turn off when receiving the shutdown trigger instruction.
9. The whole cabinet server according to claim 1, characterized in that: The computing node also includes a fan; The baseboard management controller is used to obtain and output the temperature data of the computing node where it is located; The chassis management controller is also used to obtain temperature data of each of the computing nodes through the network switching chip, and control the rotation speed of the fan of each computing node according to a preset heat dissipation strategy.
10. The whole cabinet server according to claim 1, characterized in that: The auxiliary management module is also used to configure the management switch chip to a memory mapping mode to obtain the type, connection rate and connection bandwidth of each port of the management switch chip.
11. The whole cabinet server according to claim 1, characterized in that: The baseboard management controller is also used to set a static Internet Protocol address according to its own independent identification information, so that the chassis management controller can access the corresponding computing node according to the static Internet Protocol address.
12. The whole cabinet server according to claim 1, characterized in that: The memory device is a double-rate memory stick or a memory expansion card.
13. The whole cabinet server according to claim 1, characterized in that: The auxiliary management module is a microprocessor or a baseboard management controller migration module.
14. The whole cabinet server according to any one of claims 1 to 13, characterized in that: There are multiple network switching chips, including an internal network switching chip and an external network switching chip, the external network port of the chassis management controller is connected to the external network switching chip, the internal network port of the chassis management controller is connected to the internal network switching chip, the baseboard management controller is connected to the internal network switching chip, and the external management device is connected to the external network switching chip; The external management device is used to output out-of-band management instructions; The chassis management controller is further configured to manage the computing node according to the out-of-band management instruction.
