A computing resource aggregation method and system based on pooled memory

CN118519753BActive Publication Date: 2024-11-26CHINA TELECOM CLOUD TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410985694.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-23
Publication Date
2024-11-26
Estimated Expiration
2044-07-23

AI Technical Summary

Technical Problem

In the existing cloud computing technology, the virtual resources of cloud hosts are limited by the physical resources of the computing nodes and the memory wall of the CPU wall, resulting in low resource utilization, physical resource specification bottlenecks occur in vertical amplification, and horizontal amplification increases programming complexity and Difficulty.

Method used

By building a shared memory resource pool and CPU resource pool based on the bus Internet network protocol, resource aggregation between computing nodes is realized, physical resource limitations are avoided, and programming mode is optimized through interrupt forwarding and affinity scheduling.

Benefits of technology

It significantly increases the virtual resources of cloud hosts, reduces programming complexity and difficulty, avoids the bottleneck of physical resource specifications of a single computing node, and improves resource utilization and system performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118519753B_ABST
    Figure CN118519753B_ABST
Patent Text Reader

Abstract

The present invention relates to a computing resource aggregation method and system based on pooled memory, which belongs to the field of cloud computing resource aggregation, including: obtaining computing resources of multiple computing nodes, the computing resources including memory resources and CPU resources; constructing a shared memory resource pool interconnected between each computing node based on a bus interconnection network protocol; aggregating the memory resources through the shared memory resource pool; dividing management areas in the shared memory resource pool; constructing a fast channel for data transmission within the pool in the management area; according to the fast channel, constructing a shared CPU resource pool interconnected between each computing node through interrupt forwarding and affinity scheduling; aggregating the CPU resources through the shared CPU resource pool. The present invention can significantly increase the specifications of cloud hosts, greatly increase the virtual resources of cloud hosts, and at the same time, expand the content of computing nodes, reducing the complexity and difficulty of programming modes.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of cloud computing resource aggregation, and in particular relates to a computing resource aggregation method and system based on pooled memory. Background Art

[0002] In the current cloud computing environment, virtualization technology is its cornerstone and foundation. Through virtualization technology, the physical resources of computing nodes can be abstracted into virtual resources with similar performance and complete functions. The cloud host (Virtual Machine, VM) runs on these virtual resources and can be regarded as an efficient and isolated replica of the real physical machine. This technology enables data centers to use physical resources more efficiently while providing good isolation and security.

[0003] In the existing technology, in order to rationally utilize data center computing resources and improve overall utilization, cloud computing vendors usually adopt a "one virtual many" approach to abstract one computing node into multiple cloud hosts.

[0004] In addition, with the rise of scenarios such as big data mining, AI model training, and chip simulation, CPU-intensive and memory-intensive businesses are increasingly constrained by the specifications of computing nodes and cloud hosts. For big data application scenarios, the industry usually uses two methods to expand computing resources to provide computing power: vertical expansion (Scale Up) and horizontal expansion (Scale Out). Vertical expansion aims to increase the physical resource specifications of a single computing node, while horizontal expansion manages multiple computing nodes in a distributed manner to meet the needs of big data processing.

[0005] In the above technical solution, a "one virtual multiple" approach is adopted to abstract one computing node into multiple cloud hosts. Limited by the physical resources of the computing nodes and the CPU and memory walls between computing nodes, the cloud hosts have fewer virtual resources available. The vertical expansion of computing resources leads to a bottleneck in the physical resource specifications of a single computing node. The horizontal expansion of computing resources increases the complexity and difficulty of programming in a distributed manner. Summary of the invention

[0006] In view of the above deficiencies in the prior art, the purpose of the invention is to provide a computing resource aggregation method and system based on pooled memory, which can significantly increase the cloud host specifications and greatly increase the virtual resources of the cloud host by constructing a shared memory resource pool interconnected between each computing node based on a bus interconnection network protocol and constructing a shared CPU resource pool interconnected between each computing node, so that the virtual resources available to the cloud host are increased, and are not limited by the physical resources of the computing nodes and the CPU walls and memory walls between the computing nodes. The content of the computing nodes is expanded to avoid bottlenecks in the physical resource specifications of a single computing node, and by avoiding a distributed manner based on the bus interconnection network protocol, the complexity and difficulty of the programming model are reduced.

[0007] A first aspect of the present invention provides a computing resource aggregation method based on pooled memory, comprising:

[0008] S1, obtaining computing resources of multiple computing nodes, wherein the computing resources include memory resources and CPU resources;

[0009] S2, based on the bus interconnection network protocol, builds a shared memory resource pool interconnected between various computing nodes;

[0010] S3, aggregating the memory resources through the shared memory resource pool;

[0011] S4, dividing management areas in the shared memory resource pool;

[0012] S5, constructing a fast channel for data transmission within the pool in the management area;

[0013] S6, building a shared CPU resource pool interconnected between computing nodes through interrupt forwarding and affinity scheduling according to the fast channel;

[0014] S7, aggregating the CPU resources through the shared CPU resource pool.

[0015] Further, after S2, the computing resource aggregation method based on pooled memory further includes:

[0016] Expand the content of the computing node through the shared memory resource pool:

[0017]

[0018] in, Indicates the total amount of memory accessed. represents the local memory of the computing node, Represents the remote memory of a compute node.

[0019] Furthermore, the S3 specifically includes:

[0020] S301, dividing the shared memory resource pool into a plurality of memory units through a PCIE switch and a memory controller;

[0021] S302, each computing node sends a remote memory application request to the shared memory resource pool through the PCIE switch and the memory controller;

[0022] S303: According to the request message of the remote memory application request, the memory controller locates the remote memory application request to some memory units of the shared memory resource pool;

[0023] S304, transparently transmitting the part of the memory unit to each computing node through the Dax device mode;

[0024] S305, through the memory access instruction, the computing node performs nanosecond level memory data reading and writing on the remote memory of the partial memory unit after transparent transmission, so as to aggregate the memory resources.

[0025] Furthermore, the S6 specifically includes:

[0026] S601, optimizing the interrupt forwarding through the task affinity scheduling;

[0027] S602, accelerating the optimized interrupt forwarding through the fast channel;

[0028] S603, through the accelerated interrupt forwarding, the interrupt request of the accelerated interrupt forwarding is transparently transmitted across nodes to the advanced programmable interrupt controller of the CPU of the designated node, so as to build a shared CPU resource pool interconnected with various computing nodes.

[0029] Furthermore, the S601 specifically includes:

[0030] S6011, allocating virtual CPU resources abstracted from each host machine to different NUMA nodes through the affinity scheduling;

[0031] S6012, setting the distance and penalty item between the NUMA nodes according to the communication distance between the computing nodes;

[0032] S6013, judging whether the CPU resource utilization pressure in the current NUMA node exceeds the threshold through the set penalty item; if so, automatically switching the virtual CPU resource to the remaining NUMA nodes; otherwise, keeping the virtual CPU resource running in the current NUMA node.

[0033] Furthermore, the S602 specifically includes:

[0034] S6021, sending the interrupt request of the interrupt forwarding to the fast channel;

[0035] S6022, writing the interrupt request into a buffer area in the fast channel;

[0036] S6023, sending an interrupt request to the listening process of the target computing node through the fast channel after completing the interrupt request writing;

[0037] S6024: According to the interrupt request, the target computing node reads the data in the cache area locally.

[0038] Further, after S7, the computing resource aggregation method based on pooled memory further includes:

[0039] S8, synchronizing the memory data between the computing nodes through the memory state transfer mechanism and active cache refresh.

[0040] Furthermore, the memory state transfer mechanism specifically includes:

[0041] Divide memory pages into initial, shared, modifiable, and invalid states;

[0042] The memory page EPT page table permissions of each computing node are modified through the initial, shared, modifiable and invalid states.

[0043] Furthermore, the active cache refresh specifically includes:

[0044] When the computing node performs nanosecond-level memory data reading and writing on the remote memory of the part of the memory unit after transparent transmission, the computing node flushes the local CPU cache to the shared memory resource pool;

[0045] After the local CPU cache is flushed into the shared memory resource pool, the memory page is removed from the local CPU cache.

[0046] A second aspect of the present invention provides a computing resource aggregation system based on pooled memory, comprising: a memory and one or more processors;

[0047] One or more application programs are stored in the memory, and the one or more application programs are suitable for being executed by the one or more processors to implement the above-mentioned computing resource aggregation method based on pooled memory.

[0048] The beneficial effects of the present invention are as follows:

[0049] The method and system of the present invention significantly increase the specifications of the cloud host and greatly increase the virtual resources of the cloud host by constructing a shared memory resource pool interconnected between each computing node based on a bus interconnection network protocol and constructing a shared CPU resource pool interconnected between each computing node, so that the virtual resources available to the cloud host are increased, and are not limited by the physical resources of the computing nodes and the CPU walls and memory walls between the computing nodes. The content of the computing nodes is expanded to avoid bottlenecks in the physical resource specifications of a single computing node, and the complexity and difficulty of the programming model are reduced by avoiding a distributed manner based on the bus interconnection network protocol. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings are only used to illustrate specific embodiments and are not considered to limit the present invention. In the entire drawings, the same reference symbols represent the same components. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0051] Figure 1 A schematic diagram of a flow chart of a computing resource aggregation method based on pooled memory provided in an embodiment of the present invention;

[0052] Figure 2 A schematic diagram of the structure of a computing resource aggregation system based on pooled memory provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. It should be understood that these descriptions are only exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making creative work should fall within the scope of protection of the present invention.

[0054] Furthermore, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily obscuring the concepts disclosed in the present invention.

[0055] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the orientation or position relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inside", "outside", etc. is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention. In addition, the terms "first", "second", and "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. The terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0056] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of methods and systems consistent with some aspects of the present invention as detailed in the appended claims.

[0057] The present invention proposes a computing resource aggregation method and system based on pooled memory, which solves the problem that the existing technology adopts the "one virtual multiple" approach to abstract a computing node into multiple cloud hosts. Due to the physical resources of the computing nodes and the CPU wall and memory wall between the computing nodes, the cloud hosts have fewer virtual resources available. When computing resources are vertically expanded, the physical resource specifications of a single computing node become bottlenecks. When computing resources are horizontally expanded, the complexity of the programming model and the difficulty of programming are increased in a distributed manner.

[0058] Method Embodiment

[0059] Reference Manual Attached Figure 1 , showing a flow chart of a computing resource aggregation method based on pooled memory provided in an embodiment of the present invention.

[0060] An embodiment of the present invention provides a computing resource aggregation method based on pooled memory, comprising:

[0061] Specifically, the method includes steps S1 to S8.

[0062] S1, obtain computing resources of multiple computing nodes, where the computing resources include memory resources and CPU resources.

[0063] S2, based on the bus interconnection network protocol, builds a shared memory resource pool that interconnects each computing node.

[0064] It should be noted that the key to the memory resource pooling function is the bus interconnection network protocol.

[0065] Optionally, the bus interconnection network protocols include CXL and UB.

[0066] Among them, CXL (Compute Express Link) is a high-speed interconnect technology designed to provide efficient data communication between CPU, memory and accelerator devices. It combines PCI Express and memory interfaces to support high-bandwidth, low-latency data transmission, and is suitable for data centers and high-performance computing fields. It can promote the integration of heterogeneous computing accelerators and optimize data mobility.

[0067] Among them, PCI Express (PCIe) is a high-speed serial computer expansion bus standard used to connect the motherboard with various external hardware devices, such as graphics cards, solid-state drives, and network interface cards. Compared with traditional parallel buses, PCIe has higher bandwidth, lower latency, and stronger scalability, and provides efficient data transmission through point-to-point bidirectional links, thereby significantly improving the overall performance and expansion capabilities of computer systems.

[0068] Among them, UB (Unified Bus) is an integrated circuit design concept that aims to integrate the communication needs of multiple functional modules into a shared bus architecture. Through the unified bus, different functional modules such as processors, memory, input and output devices, etc. can share data and control signals, thereby improving the overall efficiency and performance of the system. Unified bus technology is widely used in computer architecture to simplify hardware design, reduce costs, and optimize data flow and resource utilization.

[0069] Specifically, based on this type of bus interconnection network protocol, a data center can build a shared memory resource pool that is interconnected between various computing nodes.

[0070] Furthermore, through the shared memory resource pool, different computing nodes directly access the same part of the memory units on the memory resource pool, realizing cross-node memory sharing and interconnection, avoiding data copying and network transmission problems caused by distributed shared memory.

[0071] In the present invention, the CXL protocol combines PCIe and memory interfaces to provide high-bandwidth, low-latency data transmission. Through this high-speed interconnection technology, different computing nodes can efficiently share and access memory resources, thereby improving the performance of the overall system. Through a shared memory resource pool, computing nodes can directly access the same part of data in the shared memory, avoiding frequent data copying and network transmission in traditional distributed shared memory systems, thereby reducing communication overhead and latency. Through high-speed interconnection protocols such as PCIe and CXL, computing nodes and memory resources can be easily added to achieve horizontal expansion of the system. This is particularly important for data centers that require large-scale computing and high performance.

[0072] In a possible implementation manner, after S2, the method further includes:

[0073] By sharing the memory resource pool, the content of the computing node is expanded:

[0074]

[0075] in, Indicates the total amount of memory accessed. represents the local memory of the computing node, Represents the remote memory of a compute node.

[0076] Specifically, the total amount of memory accessed by a computing node is determined by the computing node's local memory and remote memory, and the maximum range of the remote memory capacity is equal to the capacity of the shared memory resource pool.

[0077] Furthermore, the memory resource pool is shared across nodes, and the reading and writing of memory resource pool data by computing nodes is actively perceived by other nodes to synchronize memory data across nodes.

[0078] In the present invention, by combining local memory and remote memory, computing nodes can use the remote memory in the shared memory resource pool to expand the total amount of memory they can access. This method breaks through the limitations of the physical memory of a single node and provides a larger memory space to meet applications with high memory requirements. The shared memory resource pool enables memory resources to be dynamically allocated and shared among multiple computing nodes, avoiding the waste of memory resources. For example, when the local memory of a node is insufficient, the remote memory in the shared memory pool can be used, thereby improving the resource utilization of the entire system. Memory sharing across nodes is achieved through the memory resource pool, and the computing nodes actively perceive the reading and writing of memory resource pool data, ensuring the consistency and synchronization of data between different nodes. This avoids calculation errors caused by inconsistent data and improves the reliability of the system.

[0079] S3,aggregates memory resources through a shared memory resource pool.

[0080] In a possible implementation, S3 specifically includes sub-steps S301 to S305:

[0081] S301, divide the shared memory resource pool into multiple memory units through a PCIE switch and a memory controller.

[0082] Specifically, through the PCIE switch and memory controller, the shared memory resource pool is divided into multiple memory units, and each memory unit is mapped to a different computing node as the Dax device of the computing node and the virtual memory of the cloud host.

[0083] Among them, DAX device (Direct Access) is a special device that allows applications to directly access device memory through direct memory access (DMA) without going through the operating system's buffer. This direct access method can significantly improve the efficiency and performance of data transmission, and is particularly suitable for application scenarios that require high-speed data reading and writing and low latency, such as high-performance computing and data center applications.

[0084] S302, each computing node sends a remote memory application request to the shared memory resource pool through the PCIE switch and the memory controller.

[0085] Optionally, the remote memory application request includes a memory resource pool ID, a starting memory unit address, and an applied memory length.

[0086] S303: According to the request message of the remote memory application request, the memory controller locates the remote memory application request to some memory units in the shared memory resource pool.

[0087] S304, transparently transmit part of the memory unit to each computing node through the Dax device mode.

[0088] S305, through the memory access instruction, the computing node performs nanosecond-level memory data reading and writing on the remote memory of the partial memory unit after transparent transmission, so as to aggregate the memory resources.

[0089] Specifically, within the computing node, the operating system and application programs use memory access instructions to read and write nanosecond memory data to the remote memory (Remote Memory) transmitted through the memory resource pool. While ensuring business performance, the memory resource pool greatly increases the memory capacity of the computing node and synchronizes memory data across nodes in a shared manner.

[0090] In the present invention, by dividing the shared memory resource pool into multiple memory units and allowing the computing nodes to access these memory units through Dax devices, the memory capacity of each computing node can be greatly increased. This method of expanding memory capacity is more flexible and cost-effective than traditional physical memory expansion. Allowing computing nodes to access remote memory in the shared memory resource pool, through nanosecond memory data read and write operations, can enable computing nodes to achieve fast access to large-capacity memory under high performance requirements. This operation method avoids traditional data copying and delay based on network transmission, and improves the overall operating efficiency and response speed of the system.

[0091] S4, divide the management area in the shared memory resource pool.

[0092] S5, building a fast channel for data transmission within the pool in the management area.

[0093] Specifically, the shared memory resource pool is directly accessed by each computing node, and the access latency is in nanoseconds, and the access speed is fast, so a management area is divided in the shared memory resource pool.

[0094] Furthermore, a fast channel for data transmission within the pool is constructed in the management area.

[0095] Furthermore, the CPU interrupt forwarding across nodes is accelerated through the fast channel.

[0096] In the present invention, constructing a fast channel for data transmission within the pool can significantly reduce the delay of data transmission, so that the data access speed in the shared memory resource pool reaches the nanosecond level. This is very critical for applications that require real-time response and high throughput, such as high-performance computing and large-scale data processing tasks in data centers. The fast channel accelerates the interrupt forwarding of the CPU across nodes, so that the computing nodes can respond and process interrupt requests from other nodes more quickly. This efficient data transmission mechanism can greatly shorten the time of data processing, thereby improving the overall response speed and processing power of the system. By accelerating interrupt forwarding through the fast channel, the system can more effectively handle a large number of interrupt requests, support the horizontal expansion of the system and the access of more computing nodes. This scalability enables the system to cope with the growing workload and data processing needs.

[0097] S6, based on the fast channel, through interrupt forwarding and affinity scheduling, build a shared CPU resource pool interconnected between various computing nodes.

[0098] In a possible implementation, S6 specifically includes sub-steps S601 to S603:

[0099] S601, optimizing interrupt forwarding through task affinity scheduling.

[0100] In a possible implementation, S601 specifically includes sub-steps S6011 to S6013:

[0101] S6011 allocates the virtual CPU resources abstracted from each host machine to different NUMA nodes through task affinity scheduling.

[0102] Among them, NUMA (Non-Uniform Memory Access) node is a multi-processor system architecture design, in which the access latency between different processor cores and memory modules in the system may be different. Each NUMA node contains a set of processor cores and dedicated local memory. This design can improve memory access efficiency and system performance, especially in application scenarios that process large amounts of data or require high concurrent access.

[0103] Among them, task affinity scheduling is a scheduling strategy that aims to optimize the execution efficiency of tasks on the processor. By setting the distance and penalty items between NUMA nodes, tasks (multi-threads or processes) are bound to the same NUMA node as much as possible to avoid data communication and non-local memory access between threads or processes across NUMA nodes.

[0104] S6012: Set the distance and penalty item between NUMA nodes according to the communication distance between each computing node.

[0105] S6013: Determine whether the CPU resource utilization pressure in the current NUMA node exceeds the threshold value by using the set penalty item. If so, the virtual CPU resource is automatically switched to the other NUMA nodes. Otherwise, the virtual CPU resource is kept running in the current NUMA node.

[0106] Specifically, after the virtual CPU resources abstracted from each host machine are allocated to different NUMA nodes, the virtual CPU resources remain running in the same NUMA node. Only when the CPU resource utilization pressure in the current NUMA node is too high, it will automatically switch to other NUMA nodes, reducing the interrupt forwarding frequency across nodes and improving business performance.

[0107] It should be noted that those skilled in the art can set the threshold value according to actual needs, and the present invention does not limit this.

[0108] In the present invention, virtual CPU resources are allocated to different NUMA nodes using task affinity scheduling, which can maximize the optimization of computing node resource utilization. By making tasks and physical CPU resources affinity, the delay and performance loss caused by cross-NUMA node access can be reduced, and the computing efficiency and response speed of the overall system can be improved. The distance and penalty items between NUMA nodes are set, and dynamically adjusted according to the CPU resource utilization pressure in the current NUMA node, which can reduce the interrupt forwarding frequency across nodes. This optimization reduces the communication overhead and data transmission delay between nodes, which helps to improve the efficiency and performance of business processing. By automatically adjusting the distribution of virtual CPU resources between NUMA nodes, the system can better adapt to load changes and the needs of different applications. This flexibility and adaptability improves the scalability of the system and supports the integration and expansion of more computing nodes.

[0109] S602, accelerating the optimized interrupt forwarding through the fast channel.

[0110] In a possible implementation, S602 specifically includes sub-steps S6021 to S6023:

[0111] S6021, sending the interrupt request of the interrupt forwarding to the fast channel.

[0112] Specifically, each computing node will create two processes: memory resource pool listening (Listen) and transmission (Forward). When interrupt forwarding is required, the hyper-aggregated host will send the interrupt request to the fast channel in the pool through the transmission process.

[0113] It should be noted that the interrupt request is an IPI interrupt request.

[0114] Among them, IPI (Inter-Processor Interrupt) is an interrupt request used to implement inter-processor communication in a multi-processor system. IPI interrupt request allows one processor to send an interrupt signal to another processor to coordinate tasks, synchronize data or notify specific events, thereby improving the system's parallel processing capabilities and efficiency.

[0115] S6022, write the interrupt request into the buffer area in the fast channel.

[0116] S6023, sending an interrupt request to the monitoring process of the target computing node through the fast channel after completing the interrupt request writing.

[0117] Specifically, when the cache data is copied, the fast channel in the pool sends an IO interrupt to the original computing node, indicating that the requested data has been written. At the same time, the fast channel in the pool sends an interrupt request to the listening process of the target computing node.

[0118] S6024: According to the interrupt request, the target computing node reads the data in the cache area locally.

[0119] Specifically, the target computing node is notified to read the designated buffer data locally, thereby realizing the process of sending an IPI interrupt from the original computing node to the target computing node.

[0120] In the present invention, the use of fast channels to accelerate interrupt forwarding can significantly reduce the transmission delay and response time of interrupt requests. This is very critical for applications and tasks that require real-time response, especially in multi-processor systems. Rapid transmission of interrupt requests through fast channels can speed up the scheduling and execution of tasks. The rapid transmission and processing of interrupt requests enhances the parallel processing capabilities of the system. In a multi-processor system, fast communication and coordination between processors can more efficiently execute tasks and improve the overall parallel computing capabilities and efficiency of the system. The optimized interrupt forwarding acceleration can reduce the load of system interrupt processing, especially in high-load periods. By processing interrupt requests more quickly, the burden on the system can be effectively reduced and the stability and performance of the system can be maintained.

[0121] S603, through the accelerated interrupt forwarding, the interrupt request of the accelerated interrupt forwarding is transparently transmitted across nodes to the advanced programmable interrupt controller of the CPU of the designated node, so as to build a shared CPU resource pool interconnected with various computing nodes.

[0122] It should be noted that in each computing node, the virtual CPU abstracted from the local physical CPU resources is called the local CPU resources, while the virtual CPUs provided by other computing nodes are called remote CPU resources. In order to reasonably utilize the remote CPU resources across nodes, the virtual CPU resources provided by multiple computing nodes are integrated into one host through interrupt forwarding for user use.

[0123] Specifically, using the interrupt forwarding method, the IPI interrupt is directly transmitted across nodes to the Advanced Programmable Interrupt Controller (APIC) of the CPU of the specified node through network transmission or the fast path within the pool, thus realizing process communication between CPUs across nodes and CPU resource utilization across nodes.

[0124] In the present invention, the interrupt forwarding technology enables the remote virtual CPU resources provided by multiple computing nodes to be integrated into one host for use by local users. This method effectively expands the computing power and resource utilization of the system, makes full use of the virtual CPU resources distributed in different nodes, and improves the computing performance and efficiency of the overall system. The IPI interrupt is directly transmitted across nodes to the APIC of the specified node through interrupt forwarding, avoiding traditional data copying and network transmission overhead. This optimization reduces the delay and bandwidth consumption of the system during cross-node communication, and improves the efficiency and real-time performance of data transmission. Interrupt forwarding technology realizes process communication and resource utilization between CPUs across nodes, and can more efficiently coordinate and process complex tasks and distributed computing. Applications and services within the system can achieve rapid data exchange and sharing in this way, further improving the parallel processing capability and overall performance of the system.

[0125] S7, aggregates CPU resources by sharing the CPU resource pool.

[0126] In a possible implementation manner, after S7, the method further includes:

[0127] S8 synchronizes memory data between computing nodes through memory state transfer mechanism and active cache refresh.

[0128] The memory state transfer mechanism specifically includes:

[0129] Divide memory pages into initial, shared, modifiable, and invalid states.

[0130] Specifically, the memory page is initially in the initialization state. The hyper-converged host's reading and writing of the memory page in the initial state will trigger a page fault and fall into KVM, and modify the memory page state to a shared state or a modifiable state. In the shared state, the local write operation of the memory page will convert the memory page state to a modifiable state. When other remote nodes need to write the same memory page data in the memory resource pool, the local node memory page state needs to be set to invalid. As a result, the next local read and write operation will trigger a page permission fault, request the latest memory data from the remote node, and then perform read and write operations on the memory data, and modify the local and remote memory page permissions. If the remote node needs to read the memory page data, when the local node state is a modifiable state, the memory page permissions will be modified to a shared state to prevent the local node from modifying the memory page without notifying the remote node.

[0131] Modify the memory page EPT page table permissions of each computing node through initial, shared, modifiable, and invalid states.

[0132] Specifically, through the flow between different states, the hyper-aggregated host modifies the EPT page table permissions of each node's memory page to achieve synchronous communication of memory data between nodes.

[0133] The active cache refresh specifically includes:

[0134] When the computing node performs nanosecond memory data reading and writing to the remote memory of some memory units after transparent transmission, it flushes the local CPU cache to the shared memory resource pool.

[0135] After flushing the local CPU cache to the shared memory resource pool, the memory page is removed from the local CPU cache.

[0136] In the present invention, the memory state transfer mechanism divides the memory pages into initial, shared, modifiable and invalid states, and manages and synchronizes the memory data through the transfer between different states. This mechanism ensures that the read and write operations of the memory data can maintain consistency in the multi-node shared memory resource pool, avoiding data conflicts and inconsistencies. The memory state transfer mechanism can minimize the communication overhead between nodes by dynamically adjusting and managing the memory page states. Data synchronization is triggered only when necessary, such as requesting the latest data from the remote node when the memory page of the local node fails, avoiding frequent data transmission, thereby improving the overall efficiency and performance of the system. The memory state transfer mechanism combines active cache refresh to support hyper-aggregated hosts to dynamically adjust and manage data in the memory resource pool. This flexibility and scalability enables the system to automatically optimize resource allocation according to different workloads and requirements, improving the responsiveness and adaptability of the system.

[0137] In a possible implementation, the resource aggregation method based on pooled memory further includes:

[0138] A hyper-converged host is built by sharing memory resource pools and CPU resource pools.

[0139] Specifically, the super-aggregated host uses a shared memory pool based on a new bus interconnection network as its hardware base. Multiple computing nodes are connected to the same memory block of the memory pool through a new bus interconnection network protocol to achieve the expansion of the memory scale of the computing nodes and the synchronization of memory data across nodes. As for the CPU resources of the computing nodes, the super-aggregated host forms a shared CPU resource pool in the form of interrupt forwarding, and cooperates with CPU affinity scheduling to avoid cross-node scheduling of processes. With the new distributed virtualization and kernel technology, the computing resources of the CPU resource pool and the memory pool are abstracted into virtual resources and provided to the super-aggregated host, realizing cross-node resource aggregation and breaking the resource isolation between computing nodes. Inside the super-aggregated host, customers can enjoy ultra-large-scale computing resources while avoiding complex distributed synchronization logic and running upper-level business applications without perception.

[0140] Through the hyper-aggregated host, the physical computing resources in multiple computing nodes are aggregated, the underlying computing node topology relationship and data synchronization mechanism are shielded, and abstracted into a unified super-large-scale host in a distributed virtualized manner.

[0141] Furthermore, customers can run their services directly on the hyper-converged host without having to consider the distributed framework.

[0142] In the present invention, the hyper-aggregated host connects the computing nodes to the shared memory resource pool through a new bus interconnection network protocol, thereby achieving the expansion of memory scale and synchronization of memory data across nodes. This architecture simplifies the complexity of system management and deployment, reduces management costs, and improves the maintainability and scalability of the system. The hyper-aggregated host abstracts the CPU and memory resources into unified virtual resources through distributed virtualization and kernel technology. This virtualization method allows customers to flexibly configure and adjust computing resources, dynamically allocate resources according to actual needs, and improve the responsiveness and adaptability of the system. The hyper-aggregated host avoids cross-node scheduling of processes and optimizes the performance and stability of the system through technologies such as CPU affinity scheduling and interrupt forwarding. At the same time, for application scenarios with high requirements for data synchronization and security, it can provide reliable protection to ensure data consistency and integrity.

[0143] System Example

[0144] Reference Manual Attached Figure 2 , showing a structural diagram of a computing resource aggregation system based on pooled memory provided in an embodiment of the present invention.

[0145] The present invention proposes a computing resource aggregation system 30 based on pooled memory, including: a memory 303 and one or more processors 301.

[0146] One or more application programs are stored in the memory 303 , and the one or more application programs are suitable for being executed by the one or more processors 301 to implement the computing resource aggregation method based on pooled memory described in the method embodiment.

[0147] The computing resource aggregation system 30 based on pooled memory includes: a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, for example, through a bus 302.

[0148] The structure of the computing resource aggregation system 30 based on pooled memory does not constitute a limitation on the embodiment of the present invention.

[0149] Processor 301 may be a CPU, a general purpose processor, a DSP, an ASIC, an FPGA or other programmable logic device, a transistor logic device, a hardware component or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of the present invention. Processor 301 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0150] The bus 302 may include a path to transmit information between the above components. The bus 302 may be a PCI bus or an EISA bus, etc. The bus 302 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0151] The memory 303 can be a ROM or other type of static storage device that can store static information and instructions, a RAM or other type of dynamic storage device that can store information and instructions, or an EEPROM, a CD-ROM or other optical disk storage, an optical disk storage (including a compressed optical disk, a laser disk, an optical disk, a digital versatile disk, a Blu-ray disk, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited to these.

[0152] Computer Readable Storage Medium Embodiments

[0153] The present invention proposes a computer-readable storage medium having a computer program stored thereon, wherein the computer program can be loaded and executed by a processor according to the computing resource aggregation method based on pooled memory as described in the first aspect.

[0154] The applicant of the present invention has made a detailed explanation and description of the implementation examples of the present invention in conjunction with the drawings in the specification. However, those skilled in the art should understand that the above implementation examples are only preferred implementation schemes of the present invention, and the detailed description is only to help readers better understand the spirit of the present invention, but not to limit the scope of protection of the present invention. On the contrary, any improvements or modifications based on the inventive spirit of the present invention should fall within the scope of protection of the present invention.

[0155] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the above embodiments, or replace some of the technical features therein by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or replacements that can be easily thought of by those skilled in the art within the technical scope disclosed in the present invention should be covered within the protection scope of the present invention.

Claims

1. A computing resource aggregation method based on pooled memory, characterized in that: include: S1, obtaining computing resources of multiple computing nodes, wherein the computing resources include memory resources and CPU resources; S2, based on the bus interconnection network protocol, builds a shared memory resource pool interconnected between various computing nodes; S3, aggregating the memory resources through the shared memory resource pool; S4, dividing management areas in the shared memory resource pool; S5, constructing a fast channel for data transmission within the pool in the management area; S6, building a shared CPU resource pool interconnected between computing nodes through interrupt forwarding and affinity scheduling according to the fast channel; S7, aggregating the CPU resources through the shared CPU resource pool; Wherein, the S3 specifically includes: S301, dividing the shared memory resource pool into a plurality of memory units through a PCIE switch and a memory controller; S302, each computing node sends a remote memory application request to the shared memory resource pool through the PCIE switch and the memory controller; S303: According to the request message of the remote memory application request, the memory controller locates the remote memory application request to some memory units of the shared memory resource pool; S304, transparently transmitting the part of the memory unit to each computing node through the Dax device mode; S305, through the memory access instruction, the computing node performs nanosecond-level memory data reading and writing on the remote memory of the part of the memory unit after transparent transmission, so as to aggregate the memory resources; Wherein, the S6 specifically includes: S601, optimizing the interrupt forwarding through task affinity scheduling; S602, accelerating the optimized interrupt forwarding through the fast channel; S603, through the accelerated interrupt forwarding, the interrupt request of the accelerated interrupt forwarding is transparently transmitted across nodes to the advanced programmable interrupt controller of the CPU of the designated node, so as to build a shared CPU resource pool interconnected with various computing nodes.

2. The computing resource aggregation method based on pooled memory according to claim 1, characterized in that: After S2, it also includes: Expanding the content of the computing node through the shared memory resource pool: in, Indicates the total amount of memory accessed. Represents the local memory of the computing node, Represents the remote memory of a compute node.

3. The computing resource aggregation method based on pooled memory according to claim 1, characterized in that: The S601 specifically includes: S6011, allocating virtual CPU resources abstracted from each host machine to different NUMA nodes through the affinity scheduling; S6012, setting the distance and penalty item between the NUMA nodes according to the communication distance between the computing nodes; S6013, judging whether the CPU resource utilization pressure in the current NUMA node exceeds the threshold through the set penalty item; if so, automatically switching the virtual CPU resource to the remaining NUMA nodes; otherwise, keeping the virtual CPU resource running in the current NUMA node.

4. The computing resource aggregation method based on pooled memory according to claim 1, characterized in that: The S602 specifically includes: S6021, sending the interrupt request of the interrupt forwarding to the fast channel; S6022, writing the interrupt request into a buffer area in the fast channel; S6023, sending an interrupt request to the listening process of the target computing node through the fast channel after completing the interrupt request writing; S6024: According to the interrupt request, the target computing node reads the data in the cache area locally.

5. The computing resource aggregation method based on pooled memory according to claim 1, characterized in that: After S7, the method further includes: S8, synchronize the memory data between the computing nodes through the memory state transfer mechanism and active cache refresh.

6. The computing resource aggregation method based on pooled memory according to claim 5, characterized in that: The memory state transfer mechanism specifically includes: Divide memory pages into initial, shared, modifiable, and invalid states; The memory page EPT page table permissions of each computing node are modified through the initial, shared, modifiable and invalid states.

7. The computing resource aggregation method based on pooled memory according to claim 5, characterized in that: The active cache refresh specifically includes: When the computing node performs nanosecond-level memory data reading and writing on the remote memory of the part of the memory unit after transparent transmission, the local CPU cache is flushed to the shared memory resource pool; After the local CPU cache is flushed into the shared memory resource pool, the memory page is removed from the local CPU cache.

8. A computing resource aggregation system based on pooled memory, characterized in that: include: memory and one or more processors; One or more applications are stored in the memory, and the one or more applications are suitable for being executed by the one or more processors to implement the computing resource aggregation method based on pooled memory as described in any one of claims 1 to 7.