Computing power resource distributed conduction method based on ARP (Address Resolution Protocol)

Abstract

The invention discloses a computing power resource distributed conduction method based on an ARP protocol, and the method comprises the steps: transmitting the application information of a computing power resource and the release information of the computing power resource through employing a reserved field in a request message of the ARP protocol; the release information of the computing power resources explains the release condition of the computing power resources in a standardized encoding mode of the reserved fields, after the router receives the release information of the computing power resources, a cost-end computing power resource release table and application information of the computing power resources are generated, and the application information of the computing power resources is generated in a standardized encoding mode of the reserved fields. The computing power resource application condition is explained, and the router generates a cost-end computing power resource application form after receiving the computing power resource application information. According to the reserved field transmission computing power resource information of the extended ARP protocol, a new protocol does not need to be introduced or a network architecture does not need to be greatly modified, implementation complexity and invasiveness to an existing network are reduced, deployment cost is low, and popularization is easy.

Description

Technical Field

[0001] The present invention relates to the field of communication technologies, and particularly to a distributed conduction method for computing power resources based on the ARP protocol. Background Art

[0002] ARP (Address Resolution Protocol) is a protocol that resolves an IP address into an Ethernet MAC address. In a network, when a host or other network device has data to send to another host or device, it must know the network layer IP address of the other party. Since an IP datagram must be encapsulated into a frame to be sent through a physical network, the sending station must also have the physical address of the receiving station. Therefore, a mapping from the IP address to the physical address is required. ARP is the protocol that implements this function.

[0003] For computing power resource scheduling technology, computing power resource scheduling refers to the rational allocation of computing power resources so that computing tasks can be completed efficiently and orderly. Currently, a centralized resource scheduling method is mostly adopted, and the implementation mechanism mainly relies on a centralized management node to be responsible for the resource allocation and task scheduling of the entire system. To avoid the problem that the entire system will be affected due to a failure of the central management node, a distributed method is adopted to deploy the management node. For example, multiple management centers are distributedly deployed, but in essence, it is still a centralized method, and the problem of single-point failure has not been completely solved. Summary of the Invention

[0004] Therefore, the present invention provides a distributed conduction method for computing power resources based on the ARP protocol to solve the problems in the prior art.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A distributed conduction method for computing power resources based on the ARP protocol transmits the application information of computing power resources and the publishing information of computing power resources by using the reserved fields in the request message of the ARP protocol.

[0007] Furthermore: The publishing information of the computing power resources describes the publishing situation of the computing power resources by means of standardized encoding of the reserved fields. After receiving the publishing information of the computing power resources, the router generates a local computing power resource publishing table and publishes the computing power resource situation to a specified router or publishes the computing power resource situation to the entire network by broadcasting, so as to realize the information synchronization between the local computing power resources and the remote computing power resources.

[0008] Further: The application information of the computing power resources describes the application situation of the computing power resources in the way of standardizing and encoding the reserved fields. After receiving the application information of the computing power resources, the router generates a local computing power resources application form, and applies for computing power resources to the designated router or broadcasts the application situation of the computing power resources to the whole network to realize the application of the local computing power resources.

[0009] Further: The reserved fields are encoded into a computing power resource application type and a computing power resource publishing type according to different uses, which respectively represent the business's demand for computing power resources and the situation of the local computing power resources.

[0010] Further: The router needs to maintain two tables, namely, the computing power resources application form and the local computing power resources table. The computing power resources application form is used to summarize the application situation of the local business for computing power resources, and the local computing power resources table is used to summarize the situation of the local computing power resources.

[0011] Still further: By specifying the peer address to establish a channel for information synchronization, the computing power application information and the computing power publishing information are synchronized; in the case of no specified peer, the application and publishing of the computing power resources and other information are broadcast to the whole network

[0012] The present invention has the following advantages: Through the distributed conduction mechanism based on the ARP protocol, the single-point failure problem of the traditional centralized scheduling is completely solved, decentralized management is realized, and the fault tolerance ability and overall reliability of the system are improved; By extending the reserved fields of the ARP protocol to transmit the computing power resources information, there is no need to introduce a new protocol or significantly modify the network architecture, reducing the implementation complexity and the intrusion into the existing network, with low deployment cost and easy promotion.

[0013] Other features and advantages of the present invention will be described in the subsequent specification, and, in part, will be obvious from the specification or will be understood by implementing the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more intuitively illustrate the prior art and the present application, the following exemplary drawings are given. It should be understood that the specific shapes and structures shown in the drawings generally should not be regarded as limiting conditions when implementing the present application; for example, those skilled in the art are capable of making routine adjustments or further optimizations to the addition / deletion / attribution division of certain units (components), specific shapes, positional relationships, connection methods, dimensional proportional relationships, etc. based on the technical concept disclosed in the present application and the exemplary drawings.

[0015] Figure 1 It is a distributed conduction block diagram of a distributed conduction method for computing power resources based on the ARP protocol provided by an embodiment of the present application.

[0016] Figure 2This is a schematic diagram of the structure of the computing power resource description field in the embodiments of the present invention.

[0017] Figure 3 is Figure 2 a schematic diagram of the structure of the type field in

[0018] Figure 4 is Figure 2 a schematic diagram of the structure of the serial number field in

[0019] Figure 5 is Figure 2 a schematic diagram of the structure of the GPU field in

[0020] Figure 6 is Figure 2 a schematic diagram of the structure of the CPU field in

[0021] Figure 7 is Figure 2 a schematic diagram of the structure of the memory field in

[0022] Figure 8 is Figure 2 a schematic diagram of the structure of the storage field in

[0023] Figure 9 is Figure 2 a schematic diagram of the structure of the network field in Detailed implementation manners

[0024] The following specific embodiments illustrate the implementation manners of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. 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 embodiments are only for further explaining the present invention and should not be construed as limiting the protection scope of the present invention. Technical engineers in this field can make some non-essential improvements and adjustments to the present invention according to the above-mentioned content of the invention; based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the protection scope of the present invention.

[0025] Please refer to Figure 1 , the router is used to connect the user network or the network of computing power resources; the server is the provider or user of the computing power resources.

[0026] Each router has a "computing power resource table" and a "computing power resource requirement table";

[0027] Among them, the computing power resource table describes the computing power resource situation of the network responsible for by this router; the computing power resource requirement table describes the computing power resource requirement situation of the network responsible for by this router.

[0028] Each server has a "computing power resource table" and a "computing power resource demand table".

[0029] Among them, the computing power resource table describes the computing power resource situation of the server; the computing power resource demand table describes the computing power resource demand situation of the server.

[0030] For the convenience of elaboration, the following is set: Router A is connected to the computing power resource demand side, that is, server A1, and Router B is connected to the computing power resource provider, that is, server B1. The interconnected addresses of Router A and server A1 are 1.1.1.1 and 1.1.1.2 respectively, and the MAC addresses are 1-1-1-1 and 1-1-1-2. The interconnected addresses of Router B and server B1 are 2.1.1.1 and 2.1.1.2 respectively, and the MAC addresses are 2-1-1-1 and 2-1-1-2.

[0031] A distributed computing power resource conduction method based on the ARP protocol includes:

[0032] S1. Optimize the existing ARP protocol

[0033] (1) Current ARP implementation process:

[0034] 1) Server A1 first checks its own ARP table to determine whether it contains an ARP entry corresponding to Router A; if the corresponding MAC address is found, server A1 directly uses the MAC address in the ARP table to encapsulate the IP datagram and sends the IP datagram to Router A.

[0035] 2) If server A1 cannot find the corresponding MAC address in the ARP table, it will cache the IP datagram and then send an ARP request message in broadcast mode; the source IP address and source MAC address in the ARP request message are the IP address and MAC address of server A1, and the destination IP address and destination MAC address are the IP address of Router A and the all-zero MAC address; since the ARP request message is sent in broadcast mode, all hosts on this network segment can receive the request, but only the requested Router A will process the request.

[0036] 3) Router A compares its own IP address with the destination IP address in the ARP request message. When the two are the same, the following processing is performed: Store the IP address and MAC address of server A1 in the ARP request message into its own ARP table; then send an ARP response message to server A1 in unicast mode, which contains its own MAC address.

[0037] 4) After Server A1 receives the ARP response packet, it adds the MAC address of Router A to its own ARP table for subsequent packet forwarding. At the same time, it encapsulates the IP datagram and sends it out.

[0038] (2) Expand the meaning of the "destination MAC" in the ARP request packet to the "computing power resource request" field and the "computing power resource description"; specifically as follows:

[0039] 1) After Server A1 learns the MAC address of Router A, it periodically publishes an ARP request report with the target MAC address field being a MAC address of all 0s. After optimization, this field represents the description of the computing power resource requirements of Server A1.

[0040] As Figure 2 shown, the computing power resource description field in the ARP request packet includes: type, serial number, GPU, CPU, memory, storage, and network.

[0041] As Figure 3 shown, the type field has a length of 1 bit and is used to represent the type of the computing power resource description field; where 0 represents resource application and 1 represents resource publication.

[0042] As Figure 4 shown, the serial number field has a length of 7 bits and is used to represent the serial number of the computing power resource packet, including the message serial number and the resource serial number;

[0043] The message serial number field has a length of 1 bit and indicates whether it is the same resource application; the resource serial number field has a length of 5 bits and represents the content of the applied resource;

[0044] When the message serial numbers are the same and the resource serial numbers are the same, it represents a repeated application;

[0045] When the message serial numbers are the same and the resource serial numbers are different, it represents an accumulative application and the resources need to be superimposed;

[0046] When the message serial numbers are different, it represents different applications.

[0047] As Figure 5 shown, the GPU field has a length of 20 bits and is used to represent the information of the GPU in the computing power resource; the GPU field includes core type, floating-point operation, video memory capacity, video memory bandwidth, video memory bit width, number of cores, core frequency, acceleration frequency, and architecture.

[0048] The core type field has a length of 2 bits and is used to represent the type of the core. Among them, 01 represents a CUDA core, which is a hardware unit that provides computing power for the GPU. The more the number, the stronger the parallel processing ability, and it is suitable for large-scale parallel computing. 10 represents a Tensor core, which is a computing unit designed specifically for deep learning operations and can efficiently execute matrix multiplication operations, greatly improving the performance of deep learning calculations.

[0049] The floating-point operation field has a length of 2 bits and is used to represent the 32-bit floating-point computing power of the GPU. The field length is 2 bits. Among them, 01 represents 10 TFLOPS, 10 represents 20 TFLOPS, and 11 represents 50 TFLOPS.

[0050] The video memory capacity field has a length of 2 bits. Among them, 01 represents 1G, 01 represents 2G, and 10 represents 5G.

[0051] The video memory bandwidth field has a length of 2 bits. Among them, 01 represents 1 GB / s, 10 represents 2 GB / s, and 10 represents 5 GB / s.

[0052] The video memory bit width field has a length of [missing value in the original]. Among them, 01 represents 64 bits, 10 represents 128 bits, and 11 represents 256 bits.

[0053] The core number field has a length of 2 bits and is used to represent the number of cores of the GPU. Among them, 01 represents 1000 cores, 01 represents 2000 cores, and 10 represents 5000 cores.

[0054] The core frequency field has a length of 2 bits and is used to represent the size of the core frequency of the GPU. Among them, 01 represents 100 MHz, 10 represents 200 MHz, and 11 represents 500 MHz.

[0055] The boost frequency field has a length of 2 bits and is used to represent the size of the boost frequency of the GPU. Among them, 01 represents 0.1 GHz, 10 represents 0.5 GHz, and 11 represents 1 GHz.

[0056] The architecture field has a length of 2 bits and is used to represent the architecture of the GPU. Among them: 01 represents the NVIDIA GPU architecture, and 10 represents the AMD GPU architecture.

[0057] Such as Figure 6 As shown, the CPU field has a length of 10 bits and is used to represent the information of the CPU in the computing power resources. The CPU field includes type, main frequency, core, process, and cache.

[0058] The type field has a length of 2 bits and is used to represent the type of the CPU. Among them, 01 represents intel, 10 represents AMD, and 11 represents ARM.

[0059] The main frequency field has a length of 2 bits and is used to represent the magnitude of the CPU frequency. Among them, 01 represents 1 GHz, 10 represents 2 GHz, and 11 represents 5 GHz;

[0060] The core field has a length of 2 bits and is used to represent the number of CPU cores. Among them, 01 represents 1, 10 represents 2, and 11 represents 5;

[0061] The process field has a length of 2 bits and is used to represent the number of CPU processes. Among them, 01 represents 1, 10 represents 2, and 11 represents 5;

[0062] The cache field has a length of 2 bits and is used to represent the type of CPU. Among them, 01 represents 1M, 10 represents 5M, and 11 represents 50M.

[0063] Such as Figure 7 As shown, the memory field has a length of 4 bits and is used to represent the information of the memory in the computing power resources; the memory field includes type and specification;

[0064] The type field has a length of 2 bits and is used to represent the type of memory. Among them, 01 represents DDR2, 10 represents DDR4, and 11 represents DDR5;

[0065] The specification field has a length of 2 bits and is used to represent the type of memory. Among them, 01 represents 1G, 10 represents 2G, and 11 represents 5G.

[0066] Such as Figure 8 As shown, the storage field has a length of 4 bits and is used to represent the information of the storage in the computing power resources, including type and specification;

[0067] The type field has a length of 2 bits and is used to represent the type of storage. Among them, 01 represents HDD, and 10 represents SSD;

[0068] The specification field has a length of 2 bits and is used to represent the type of memory. Among them, 01 represents 10G, 10 represents 20G, and 11 represents 50G.

[0069] Such as Figure 9 As shown, the network field has a length of 2 bits and includes a performance field, which is used to represent the information of the network in the computing power resources; among them, the performance field has a length of 2 bits and is used to represent the type of network network interface. Among them, 01 represents 1G, 01 represents 10G, and 11 represents 50G.

[0070] (3) Expand the interaction process of ARP;

[0071] After router A receives an ARP request, it checks the "destination MAC" field:

[0072] When it is all "0", it indicates an unextended ARP packet, and it enters the ARP process;

[0073] When it is not all "0", it indicates an extended ARP packet, and the information at the "computing power resource request" position is extracted to update the local computing power resource request description table.

[0074] S2. Update the computing power resource request table

[0075] When Server A1 receives the demand for computing power resources at the application layer, it first splits the demand for computing power resources; according to the classification and structure of the resource description fields, it is split into information that can be transmitted by the "computing power resource request" field.

[0076] For example:

[0077] The customer's CPU demand description is: an Intel CPU, a main frequency of 2GHz, 4 core CPUs, 8 processes, and a cache of 50M. For the sake of simplicity in explaining the problem, the current example does not include fields such as memory, storage, and network.

[0078] After splitting, the main part of the packet is shown in Table 1 below:

[0079] Table 1

[0080]

[0081] After Router A receives the above three packets, it integrates them and restores the demand to "an Intel CPU, a main frequency of 2GHz, 4 core CPUs, 8 processes, and a cache of 50M".

[0082] Update the computing power resource request table, as shown in Table 2 below.

[0083] Table 2

[0084]

[0085] S3. Update the computing power resource table

[0086] When Server B1 publishes computing power resources, it first splits the publication of computing power resources; according to the classification and structure of the resource description fields, it is split into information that can be transmitted by the "computing power resource publication" field.

[0087] For example:

[0088] The CPU resource description of Server B1 is: an Intel CPU, a main frequency of 2GHz, 8 core CPUs, 16 processes, and a cache of 100M; for the sake of simplicity in explaining the problem, the current example does not include fields such as memory, storage, and network.

[0089] After splitting, the main part of the message is shown in Table 3 below:

[0090] Table 3

[0091]

[0092] After Router B receives the above three ARP messages, it integrates them, and restores the requirements to "Intel CPU, 2GHz main frequency, 4-core CPU, 8 processes, 50M cache", and updates the computing power resource publishing table as shown in Table 4 below:

[0093] Table 4

[0094]

[0095] After Router B restores the computing power resource publishing table, it generates the local computing power resource description table as shown in Table 5 below in combination with the received ARP messages.

[0096] Table 5

[0097]

[0098] S4. Transmit the local computing power resource description table

[0099] A connection is established between Router B, which is the router for publishing computing power resources, and Router A, which is the router for requesting computing power resources, to realize the transmission of the local computing power resource description table of Router B to Router A.

[0100] That is, Router A forms a remote computing power resource description table, as shown in Table 6 below.

[0101] Table 6

[0102]

[0103] The specific method for transmitting the local computing power resource description table is as follows:

[0104] 1) According to the computing power resource request of Server A1, Server B1 can meet the computing power resource requirements of Server A1;

[0105] 2) Router A sends a computing power resource request message to Router B. After Router B receives the computing power resource request message from Router A, it locks the computing power resources and feeds back to Router A that the locking of the computing power resources is successful;

[0106] 3) When the computing power resources of Router B are dynamic, the computing power resources may be preempted by other demanders, and the locking of the computing power resources may fail. At this time, it feeds back to Router A that the locking of the computing power resources fails;

[0107] 4) After Router A receives the failure notice of Router B's locking of computing power resources, it continues to search for computing power resources in the remote computing power resource description table of Router A. When there are no computing power resources that meet the requirements of the computing power resource request, the computing power resource request of Server A1 fails;

[0108] 5) After Router A receives the success notice of Router B's locking of computing power resources, it notifies Server A1 that it can send a computing power resource request to "destination IP address 1.1.1.2, destination MAC address 1-1-1-2". After Server B1 receives the request, it provides computing power resources to Server A1;

[0109] 6) After Server A1 finishes using the computing power resources of Server B1 and sends a request indicating that the computing power resources have been used up to Server B1, after Server B1 receives the request indicating that the computing power resources of Server A1 have been used up, it releases the computing power resources used by Server A1 and notifies Router B to update the computing power resources.

[0110] In this embodiment, currently, the existing ARP protocol is extended to implement the transfer of computing power resources. It is also possible to design a new protocol for the transfer of computing power resources to time the dissemination of computing power resources and computing power resource request information.

[0111] The ARP packets for computing power resource requests and the ARP packets for computing power resource announcements can be sent regularly. If the sending frequency is high, the computing power resources and computing power resource request information will be updated more timely, and the local computing power resource description table and the remote computing power resource description table will be more accurate, but it will occupy more bandwidth and CPU resources of the router; conversely, if the sending frequency is low, the update of the computing power resources and computing power resource request information will be untimely, and the local computing power resource description table and the remote computing power resource description table will be less accurate, but it will occupy less bandwidth and CPU resources of the router.

[0112] The local router and the remote router can establish a link for computing power resource update and request. The link can be established by manually setting the addresses of the opposite ends at both endpoints; it can also be established automatically through broadcasting, that is, the computing power resource end broadcasts the computing power resource situation, and the computing power resource request end broadcasts the computing power resource request situation. When a computing request is satisfied at a certain router, a dynamic link will be established with the computing power resource request end to complete the process of locking, using, and releasing the computing power resources.

[0113] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A distributed conduction method for computing power resources based on the ARP protocol, characterized in that, By using the reserved field in the request message of the ARP protocol, the application information of computing power resources and the release information of computing power resources are transmitted.

2. The distributed conduction method of computing power resources based on the ARP protocol according to claim 1, characterized in that The release information of the computing power resources describes the release situation of the computing power resources by means of standardized encoding of the reserved field. After receiving the release information of the computing power resources, the router generates a local computing power resource release table and publishes the computing power resource situation to the specified router or broadcasts the computing power resource situation to the whole network to realize the information synchronization between the local computing power resources and the remote computing power resources.

3. A distributed conduction method for computing power resources based on the ARP protocol according to claim 1, characterized in that The application information of the computing power resources describes the application situation of the computing power resources by means of standardized encoding of the reserved field. After receiving the application information of the computing power resources, the router generates a local computing power resource application form and applies for computing power resources to the specified router or broadcasts the application situation of the computing power resources to the whole network to realize the application of the local computing power resources.

4. A distributed conduction method for computing power resources based on the ARP protocol according to claim 1, characterized in that The reserved field is encoded into a computing power resource application type and a computing power resource release type according to different uses, which respectively represent the business's demand for computing power resources and the situation of the local computing power resources.

5. A distributed conduction method for computing power resources based on the ARP protocol according to claim 2, characterized in that, The router needs to maintain two tables, namely, the computing power resource application form and the local computing power resource table. The computing power resource application form is used to summarize the application situation of the local business for computing power resources, and the local computing power resource table is used to summarize the situation of the local computing power resources.

6. A distributed conduction method for computing power resources based on the ARP protocol according to any one of claims 1-5, characterized in that A channel for information synchronization is established by specifying the peer address to synchronize the computing power application information and the computing power release information; in the case where no peer is specified, information such as the application of computing power resources and the release of computing power resources is broadcast to the whole network.

Images