Method and system for allocating computing tasks by wide-area intelligent computing network

Abstract

The invention discloses a calculation task allocation method and system for a wide-area intelligent computing network, and belongs to the field of intelligent computing networks, and the method comprises the steps that a data center for allocating tasks issues a resource request to a data center for all non-allocation tasks through a first multicast for a calculation unit required by a to-be-allocated calculation task; the data center which does not allocate the task calculates the number of calculation units which can be provided according to the resource request issued by the first multicast and the scheduling strategy; the data center which does not distribute the task adopts the second multicast to respond to the resource request issued by the first multicast according to the number of the computing units which can be provided; and the data center of the distributed task receives the response of the second multicast, and issues the calculation task to be distributed to the data center which has the fastest response speed and can provide enough calculation units. The method does not depend on the configuration of a wide area network and can be realized as long as the network is reachable, and the task allocation efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of intelligent computing networks, and in particular to a method and system for allocating computing tasks in a wide-area intelligent computing network. Background Technology

[0002] Computing power is allocated in units of GPUs (Graphics Processing Units) or CPUs (Central Processing Units). With the rapid development of AI (Artificial Intelligence), the demand for computing power is constantly expanding, which puts forward increasingly higher requirements for computing power scheduling. When the number of computing units required by a data center exceeds the number available locally, it may be necessary to request computing power units between multiple remote data centers.

[0003] In intelligent computing networks, the request process for computing units requires low latency and a simple, clear protocol to ensure rapid task distribution and improve the utilization rate of computing units.

[0004] In existing technologies, in an environment with multiple connected data centers (e.g., three), data centers typically need to poll each other to allocate tasks, requesting resources from the first data center. If the first data center lacks sufficient resources, it then requests resources from the second data center, and so on. During this process, if the first data center lacks resources while the second data center has available resources, a significant amount of computing time is wasted. In extreme cases, a connection interruption with the first data center can cause long and unnecessary wait times. Utilizing every second of the computing unit translates to cost savings. Summary of the Invention

[0005] This application proposes a method and system for allocating computing tasks in a wide-area intelligent computing network, which solves the problem of wasting a lot of computing time when polling between multiple remote data centers.

[0006] This application provides a method for allocating computing tasks in a wide-area intelligent computing network, including: The data center that assigns tasks computes the computing tasks to be assigned in the wide-area intelligent computing network, and sends the number of computing units required by the computing tasks to all data centers that are not assigned tasks through the first multicast. The data center for non-assigned tasks calculates the number of computing units it can provide based on the resource requests and scheduling policies issued by the first multicast. Data centers that do not assign tasks respond to resource requests sent by the first multicast using a second multicast, based on the number of computing units they can provide. Based on the response from the second multicast, the data center that assigns the tasks distributes the computing tasks to be assigned to the data center with the fastest response time and the ability to provide sufficient computing units.

[0007] A method for allocating computing tasks in a wide-area intelligent computing network further includes: After receiving a second multicast response, if the number of computing units that the data center that is assigning the task cannot provide is insufficient to meet the number of computing units required by the computing task to be assigned, the computing task to be assigned is divided according to the number of computing units that the responding non-assignment task data center can provide. The divided tasks are then sent to the responding non-assignment task data center in the form of unicast. The data center continues to wait for a new response to assign the remaining tasks. After waiting for a preset time, the data center that is assigning the task sends the resource request to all non-assignment task data centers again through the first multicast.

[0008] The resource request sent by the first multicast includes the number and specifications of computing units required for the computing task to be allocated.

[0009] The second multicast response includes: the address of the gateway of this data center and the number of computing units that can be provided.

[0010] A method for allocating computing tasks in a wide-area intelligent computing network further includes: If any data center without assigned tasks receives a response from a second multicast sent by another data center without assigned tasks before responding to the resource request sent by the first multicast, and finds that the number of computing units that can be provided in the response of the second multicast is sufficient to meet the number of computing units required by the computing task to be assigned, then this data center will no longer respond.

[0011] A method for allocating computing tasks in a wide-area intelligent computing network further includes: The data center for non-assigned tasks has a first queue and a second queue. The first queue is an idle computing unit queue, which is used to store idle computing units in this data center. The second queue is a working computing unit queue, which is used to store computing units in this data center that are currently computing. After receiving a computing task to be assigned, the data center that is not assigned a task allocates the corresponding computing unit and schedules the used computing unit from the first queue to the second queue. After completing the computing task, the data center sends the response computing result to the data center that assigned the task and puts the corresponding computing unit back from the second queue to the first queue.

[0012] A method for allocating computing tasks in a wide-area intelligent computing network further includes: If the data center that assigned the task receives a response indicating that all computation tasks have been completed, then the assignment of the computation tasks to be assigned ends; if the data center that assigned the task receives a response indicating that all computation tasks have not been completed, then the remaining computation tasks are treated as new computation tasks to be assigned, and the process returns to step S1 to send resource requests for the new computation tasks to be assigned to all data centers that are not assigned tasks via the first multicast.

[0013] The first or second multicast is constructed using a method that includes: using a network routing protocol to transmit multicast group information; transmitting multicast group information within the operator's network via IGP and PIM protocols; and using M-BGP messages to transmit multicast group information across different operators.

[0014] The first or second multicast connection can be further configured by establishing a layer 2 or layer 3 connection through a tunnel. If it is a layer 2 connection, multicast data is transmitted directly between nodes. If it is a layer 3 connection, a virtual router is used to enable the PIM protocol.

[0015] This application also provides a system for allocating computing tasks in a wide-area intelligent computing network, including: Each data center is interconnected through a gateway. Each data center includes data centers that assign tasks and data centers that do not assign tasks. The first multicast is used to allocate computing tasks and issue resource requests to the wide-area intelligent computing network. The data center that allocates the task is the multicast source of the resource request, and the data center that does not allocate the task is the receiving end of the resource request. The second multicast is used to respond to resource requests, with any data center in the non-assigned task data center as the multicast source and the data center that assigns the task as the receiving end of the response to the resource request.

[0016] A system for allocating computing tasks in a wide-area intelligent computing network further includes: the data center for non-assigned tasks has a first queue and a second queue, the first queue being an idle computing unit queue for storing idle computing units in the data center, and the second queue being a working computing unit queue for storing computing units in the data center that are currently computing.

[0017] The above-mentioned technical solution adopted in this application can achieve the following beneficial effects: This application proposes a method and system for allocating computing tasks in a wide-area intelligent computing network. This method is independent of the wide-area network configuration and can be implemented as long as the network is reachable. It utilizes multicast established in a data center for task allocation, enabling parallel task allocation and improving efficiency. Attached Figure Description

[0018] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating a method for allocating computing tasks in a wide-area intelligent computing network according to an embodiment of this application. Figure 2 This is a system principle block diagram of a wide-area intelligent computing network for allocating computing tasks, as an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0021] Example 1: This embodiment provides a method for allocating computing tasks in a wide-area intelligent computing network, such as... Figure 1 As shown, it includes: Step S1: The data center that assigns tasks calculates the computing tasks to be assigned in the wide-area intelligent computing network, and sends the computing units required by the computing tasks to be assigned to all data centers that are not assigned tasks through the first multicast. The resource request sent by the first multicast includes the number and specifications of computing units required for the computing task to be allocated.

[0022] In this embodiment, to address the problem of wasted computation time and unnecessary waiting time in existing technologies when requesting resources, a first multicast and a second multicast are used between the data center allocating tasks and the data centers not allocating tasks. Resource requests are sent to all data centers not allocating tasks via the first multicast, and the second multicast responds to these requests. This method eliminates the need for polling of data centers allocating tasks, reducing computation time and unnecessary waiting time, and improving the efficiency of task allocation. Furthermore, it achieves configuration independent of the wide area network (WAN), requiring only network connectivity. Using multicast allows requests to be sent simultaneously to all geographically dispersed data centers (i.e., data centers not allocating tasks), saving request and waiting time, especially in polling environments where a connection failure in one data center can lead to lengthy waiting times.

[0023] Step S2: The data center for non-assigned tasks calculates the number of computing units it can provide based on the resource requests and scheduling policies issued by the first multicast. In this embodiment, the scheduling strategy for the number of computing units is flexible. In this scheme, the node that responds first receives the task first. In practice, there can be more scheduling methods, such as the node with the most resources receiving the task first.

[0024] In this embodiment, the algorithm for allocating tasks is simple. The data center that allocates tasks provides the required computing units, and the remote data center (i.e., the data center that does not allocate tasks) responds based on the real-time maintenance status of the local computing units and the status of the local scheduling algorithm.

[0025] Step S3: The data center for non-assigned tasks responds to the resource requests sent by the first multicast using the second multicast based on the number of computing units it can provide; It should be noted that if any data center in the non-assigned task data center receives a response to a second multicast sent by the non-assigned task data center before responding to the first multicast resource request, and finds that the number of computing units that can be provided in the second multicast response can meet the number of computing units required by the computing task to be assigned, then this data center will not respond.

[0026] Step S4: The data center that assigns the task receives the response from the second multicast and distributes the computing task to be assigned to the data center with the fastest response speed and the ability to provide sufficient computing units.

[0027] The second multicast response includes: the address of the gateway of this data center and the number of computing units that can be provided.

[0028] A method for allocating computing tasks in a wide-area intelligent computing network further includes: Step S5: After receiving a second multicast response, if the number of computing units that can be provided by the data center that is assigning the task is insufficient to meet the number of computing units required by the computing task to be assigned, the computing task to be assigned is divided according to the number of computing units that the responding non-assignment task data center can provide. The divided task is then sent to the responding non-assignment task data center in the form of unicast. The data center continues to wait for a new response to assign the remaining task. After waiting for a preset time, the data center that is assigning the task sends the resource request to all non-assignment task data centers again through the first multicast.

[0029] A method for allocating computing tasks in a wide-area intelligent computing network further includes: Step S6: After receiving the computing task to be assigned, the data center that is not assigned a task allocates the corresponding computing unit and schedules the used computing unit from the first queue to the second queue. After completing the computing task, the data center sends the response computing result to the data center that assigned the task and puts the corresponding computing unit back from the second queue to the first queue.

[0030] The data center without assigned tasks has a first queue and a second queue. The first queue is an idle computing unit queue, which is used to store idle computing units in this data center. The second queue is a working computing unit queue, which is used to store computing units in this data center that are currently computing. A method for allocating computing tasks in a wide-area intelligent computing network further includes: Step S7: If the data center that assigned the task receives a response indicating that all calculation tasks have been completed, then the calculation task to be assigned ends; if the data center that assigned the task receives a response indicating that all calculation tasks have not been completed, then the remaining calculation tasks are treated as new calculation tasks to be assigned, and the process returns to step S1 to send the new calculation tasks to all non-assigned task data centers via the first multicast resource request.

[0031] The first or second multicast is constructed using a method that includes: using network routing protocols to transmit multicast group information; transmitting multicast group information within the operator's network via IGP (Interior Gateway Protocol) and PIM (Protocol Independent Multicast); and using M-BGP (Multiprotocol Border Gateway Protocol) messages to transmit multicast group information across different operator domains when crossing operator boundaries.

[0032] In this embodiment, a network routing protocol is used to transmit multicast group information. Within the operator's network, multicast group information is transmitted via protocols such as IGP and PIM. When crossing operator boundaries, M-BGP messages are used to achieve cross-domain multicast group information transmission. This method requires the cooperation of network operators, especially when crossing domain boundaries, which requires the cooperation of multiple network operators, making implementation quite challenging.

[0033] The method for constructing the first or second multicast also includes: establishing a Layer 2 or Layer 3 connection by establishing a tunnel. If it is a Layer 2 connection, multicast data is transmitted directly between the nodes.

[0034] In this embodiment, Layer 2 or Layer 3 connections are established through tunneling, such as GRE tunnels or L2TP. Since this method is independent of the ISP, it allows for convenient establishment of Layer 2 or Layer 3 connections between multiple nodes. For Layer 2 connections, multicast data can be directly transmitted between nodes, so tunneling is the preferred method. While broadcasting can be used for service distribution and response in Layer 2 connections, its flexibility is less than that of multicasting. Layer 3 connections require enabling protocols such as PIM on the virtual router, which is relatively complex. Layer 2 tunnels only require network connectivity and do not require further configuration. Establishing Layer 2 or Layer 3 connections through tunneling is existing technology and will not be elaborated upon further in this application.

[0035] To illustrate the specific implementation process of this embodiment in more detail, an example is provided below: Step 1: Enable network interconnection between data centers through gateways. Each data center includes data centers that assign tasks and data centers that do not assign tasks.

[0036] Step 2: Configure the data center gateway that assigns tasks as the source of the first multicast address, and the data center that does not assign tasks as the receiver of the first multicast address.

[0037] Step 3: The second multicast address is used as the address for resource requests responded to by data centers that are not assigned tasks. Any data center that is not assigned task can be a multicast source, while the data center that is assigned task is the receiver.

[0038] Step 4: When the data center that assigns the task has no computing units in its own data center, it sends a resource request on the address of the first multicast, which includes the number and specifications of the required computing units.

[0039] Because the request is sent over a multicast address, the gateway network of the data center that assigns the task will send resource requests to all data centers that do not assign tasks at once.

[0040] Step 5: Each data center without assigned tasks maintains two queues: the first queue is the idle computing unit queue, and the second queue is the working computing unit queue. Upon receiving a resource request, each data center calculates the available computing units according to its scheduling policy and responds on the second multicast address. The response includes the address of the remote data center gateway and the number of available computing units.

[0041] Step 6: If a data center that is not assigned a task finds that the total available computing resources in the response received by the second multicast address already meet the required number of computing units before responding to the address request of the first multicast, it will not respond further. Otherwise, it will send the address of its own data center and the number of available data units to the second multicast address as in Step 5.

[0042] Step 7: The data center gateway that assigns the task receives the response on the second multicast address. Based on the number of available computing units received, it assigns the task. If the number of available computing units meets the required number of computing units, it directly sends the computing data to the address of the corresponding data center. If the number of available computing units does not meet the required number of computing units, it performs data segmentation of the computing task and sends the computing data to the ID (Identification) of the remote data center, and continues to wait for the response from the next data center. Step 8: After receiving the computation data, the corresponding data center gateway allocates the corresponding computation unit and schedules the computation unit to be used from the first queue to the second queue. After completing the computation task, it responds with the computation result to the data center that allocated the task and puts the original working computation unit back from the second queue to the first queue.

[0043] Step 9: The data center that assigned the task waits for the task to be sent back. If the returned result indicates that all data has been calculated, the task ends; otherwise, the resource request process in Step 4 continues for the remaining tasks.

[0044] This embodiment provides a method for allocating computing tasks in a wide-area intelligent computing network, comprising: a data center for task allocation sending resource requests for the computing tasks to be allocated to all non-assigned task data centers via a first multicast; the non-assigned task data centers calculating the number of computing units they can provide based on the resource requests sent via the first multicast and a scheduling policy; the non-assigned task data centers responding to the resource requests sent via the first multicast using a second multicast based on the number of computing units they can provide; and the data center for task allocation receiving the response from the second multicast and distributing the computing tasks to be allocated to the data center with the fastest response speed and the ability to satisfy the computing tasks to be allocated. This embodiment is independent of wide area network configuration, can be implemented as long as the network is reachable, and improves the efficiency of task allocation.

[0045] Example 2: This application also provides a system for allocating computing tasks in a wide-area intelligent computing network, such as... Figure 2 As shown, it includes: Each data center is interconnected through a gateway. Each data center includes data centers that assign tasks and data centers that do not assign tasks. The first multicast is used to allocate computing tasks and issue resource requests to the wide-area intelligent computing network. The data center that allocates the task is the multicast source of the resource request, and the data center that does not allocate the task is the receiving end of the resource request. The second multicast is used to respond to resource requests, with any data center in the non-assigned task data center as the multicast source and the data center that assigns the task as the receiving end of the response to the resource request.

[0046] A system for allocating computing tasks in a wide-area intelligent computing network further includes: the data center for non-assigned tasks has a first queue and a second queue, the first queue being an idle computing unit queue for storing idle computing units in the data center, and the second queue being a working computing unit queue for storing computing units in the data center that are currently computing.

[0047] It should be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0048] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for allocating computing tasks in a wide-area intelligent computing network, characterized in that, include: The data center that assigns tasks computes the computing tasks to be assigned in the wide-area intelligent computing network, and sends the computing units required by the computing tasks to all data centers that are not assigned tasks through the first multicast. The data center for non-assigned tasks calculates the number of computing units it can provide based on the resource requests and scheduling policies issued by the first multicast. Data centers that do not assign tasks respond to resource requests sent by the first multicast using a second multicast, based on the number of computing units they can provide. The data center that assigns tasks receives the response from the second multicast and distributes the computing tasks to be assigned to the data center with the fastest response time and the ability to provide sufficient computing units.

2. The method for allocating computing tasks in a wide-area intelligent computing network according to claim 1, characterized in that, Also includes: After receiving a second multicast response, if the number of computing units that the data center that is assigning the task cannot provide is insufficient to meet the number of computing units required by the computing task to be assigned, the computing task to be assigned is divided according to the number of computing units that the responding non-assignment task data center can provide. The divided tasks are then sent to the responding non-assignment task data center in the form of unicast. The data center continues to wait for a new response to assign the remaining tasks. After waiting for a preset time, the data center that is assigning the task sends the resource request to all non-assignment task data centers again through the first multicast.

3. The method for allocating computing tasks in a wide-area intelligent computing network according to claim 1, characterized in that, The resource request sent by the first multicast includes: the number and specifications of computing units required for the computing task to be allocated; The second multicast response includes: the address of the gateway of this data center and the number of computing units that can be provided.

4. The method for allocating computing tasks in a wide-area intelligent computing network according to claim 1, characterized in that, Also includes: If any data center without assigned tasks receives a response from a second multicast sent by another data center without assigned tasks before responding to the resource request sent by the first multicast, and finds that the number of computing units that can be provided in the response of the second multicast is sufficient to meet the number of computing units required by the computing task to be assigned, then this data center will no longer respond.

5. The method for allocating computing tasks in a wide-area intelligent computing network according to claim 1, characterized in that, Also includes: The data center for non-assigned tasks has a first queue and a second queue. The first queue is an idle computing unit queue, which is used to store idle computing units in this data center. The second queue is a working computing unit queue, which is used to store computing units in this data center that are currently computing. After receiving a computing task to be assigned, the data center that is not assigned a task allocates the corresponding computing unit and schedules the used computing unit from the first queue to the second queue. After completing the computing task, the data center sends the response computing result to the data center that assigned the task and puts the corresponding computing unit back from the second queue to the first queue.

6. The method for allocating computing tasks in a wide-area intelligent computing network according to claim 1, characterized in that, Also includes: If the data center that assigned the task receives a response indicating that all calculation tasks have been completed, then the calculation task to be assigned will be terminated. If the data center that assigned the task receives a response indicating that not all computation tasks have been completed, then the remaining computation tasks will be treated as new computation tasks to be assigned. The process will then return to step S1 and resend the new computation tasks to all data centers that are not assigned tasks via the first multicast resource request.

7. The method for allocating computing tasks in a wide-area intelligent computing network according to claim 1, characterized in that, The first or second multicast is constructed using a method that includes: using a network routing protocol to transmit multicast group information; transmitting multicast group information within the operator's network via IGP and PIM protocols; and using M-BGP messages to transmit multicast group information across different operators.

8. The method for allocating computing tasks in a wide-area intelligent computing network according to claim 1, characterized in that, The first or second multicast connection can be further configured by establishing a layer 2 or layer 3 connection through a tunnel. If it is a layer 2 connection, multicast data is transmitted directly between nodes. If it is a layer 3 connection, a virtual router is used to enable the PIM protocol.

9. A system for allocating computing tasks in a wide-area intelligent computing network, characterized in that, include: Each data center is interconnected through a gateway. Each data center includes data centers that assign tasks and data centers that do not assign tasks. The first multicast is used to allocate computing tasks and issue resource requests to the wide-area intelligent computing network. The data center that allocates the task is the multicast source of the resource request, and the data center that does not allocate the task is the receiving end of the resource request. The second multicast is used to respond to resource requests, with any data center in the non-assigned task data center as the multicast source and the data center that assigns the task as the receiving end of the response to the resource request.

10. A system for allocating computing tasks in a wide-area intelligent computing network according to claim 9, characterized in that, Also includes: The data center for non-assigned tasks has a first queue and a second queue. The first queue is an idle computing unit queue, used to store idle computing units in the data center. The second queue is a working computing unit queue, used to store computing units in the data center that are currently computing.

Images