CN112003907A - Deterministic forwarding method for network resource demand and computational power demand thereof - Google Patents
Deterministic forwarding method for network resource demand and computational power demand thereof Download PDFInfo
- Publication number
- CN112003907A CN112003907A CN202010773478.3A CN202010773478A CN112003907A CN 112003907 A CN112003907 A CN 112003907A CN 202010773478 A CN202010773478 A CN 202010773478A CN 112003907 A CN112003907 A CN 112003907A
- Authority
- CN
- China
- Prior art keywords
- node
- probability
- matchable
- resource
- nodes
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 75
- 230000006870 function Effects 0.000 claims abstract description 32
- 238000004364 calculation method Methods 0.000 claims description 46
- 230000008569 process Effects 0.000 claims description 19
- 238000004422 calculation algorithm Methods 0.000 claims description 10
- 238000000137 annealing Methods 0.000 claims description 4
- 230000008859 change Effects 0.000 abstract description 11
- 230000008901 benefit Effects 0.000 description 9
- 230000007423 decrease Effects 0.000 description 6
- 238000012986 modification Methods 0.000 description 6
- 230000004048 modification Effects 0.000 description 6
- 238000013459 approach Methods 0.000 description 4
- 238000004590 computer program Methods 0.000 description 4
- 238000001514 detection method Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000013473 artificial intelligence Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/14—Session management
- H04L67/146—Markers for unambiguous identification of a particular session, e.g. session cookie or URL-encoding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/01—Protocols
- H04L67/10—Protocols in which an application is distributed across nodes in the network
- H04L67/1001—Protocols in which an application is distributed across nodes in the network for accessing one among a plurality of replicated servers
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/50—Network services
- H04L67/60—Scheduling or organising the servicing of application requests, e.g. requests for application data transmissions using the analysis and optimisation of the required network resources
- H04L67/63—Routing a service request depending on the request content or context
Abstract
The application discloses a network resource demand and a deterministic forwarding method of the computational power demand, wherein the deterministic forwarding method of the network resource demand comprises the following steps: receiving a resource request packet and searching for a matchable node according to the resource demand information in the resource request packet; calculating the cost value of each matched node; updating the probability tree corresponding to the matchable node by adopting an online learning method according to a predefined cost function, so that the probability of the leaf node corresponding to the matchable node with the minimum cost value and the non-leaf node passing from the root node to the leaf node is increased; starting from a root node of the probability tree, selecting a sub-node of the current node according to the probability of each sub-node of the current node until the selected node is a leaf node; and forwarding the resource request packet to the matched node corresponding to the selected leaf node. The method is based on the probability tree to carry out online learning, so that the forwarding of the resource requirements can adapt to the change of the network condition, and the efficiency of the network resource requirement forwarding is improved.
Description
Technical Field
The invention relates to the technical field of network communication, in particular to a deterministic forwarding method for network resource requirements and computing power requirements thereof.
Background
The core value of the network is to improve the efficiency, and reasonably match the resource demand and the resource providing node in the network, so that the resource demand can be met to the maximum extent, and the method is the key for improving the network performance and efficiency. Resources in the network comprise computing resources, storage resources, bandwidth resources and the like, for example, a computing service network for providing the computing resources is taken as an example, and the computing service network is presented to improve the cooperative work efficiency of end, edge and cloud three-level computing and match appropriate computing resources for computing tasks. For example, an integrated Artificial Intelligence (AI) task includes three aspects of an algorithm, an algorithm and data, wherein the algorithm depends on continuous research of academic and theoretical circles, data needs to be aggregated by a cloud platform and the like to play a role, and the algorithm is based on the same algorithm and the same cost, more data are processed in the same time, so that the algorithm is the largest variable factor for a specific AI task in a certain time. Therefore, matching the computing power requirement to the most suitable computing node to execute the computation is the key for improving the performance and efficiency of the computing power service network. However, in the past, the routing and forwarding under the IP network are separated, the routing is "smart" routing, and the forwarding is only "clumsy" forwarding, and the forwarding is performed purely according to the information of the routing table, so that the resource requirement in the network is forwarded in this way, and is difficult to match to a proper resource providing node, which affects the performance and efficiency of the network.
Disclosure of Invention
The application provides a network resource demand and a deterministic forwarding method of the calculation demand thereof, so as to improve the efficiency of forwarding the network resource demand and further improve the performance of the network.
According to a first aspect, an embodiment provides a deterministic forwarding method for computing power demand, comprising:
receiving a calculation force request packet, wherein the calculation force request packet comprises calculation force demand information; the calculation force demand information comprises calculation force requester identification, calculation task types, input data positions of calculation tasks and calculation force demand quantity;
searching a calculation power matching forwarding table according to the calculation power demand information, and obtaining available computing nodes corresponding to the calculation power demand information, wherein each item of the calculation power matching forwarding table comprises calculation power demand information, the corresponding available computing nodes and attribute information of each available computing node; the attribute information comprises computing resource tension, load balancing information, fairness information and RTT time delay;
calculating the cost value of each available computing node according to the attribute information;
updating the probability tree corresponding to the available computing node by adopting an online learning method according to a predefined cost function, so that the probability of a leaf node corresponding to the available computing node with the minimum cost value and a non-leaf node passing through the leaf node from a root node is increased; the probability tree is a binary tree, and leaf nodes of the binary tree correspond to all available computing nodes;
selecting sub-nodes of the current node from the root node of the probability tree according to the probability of each sub-node of the current node until the selected node is a leaf node;
and forwarding the computing power request packet to the available computing node corresponding to the selected leaf node.
In one embodiment, the method further includes detecting whether the available computing node has a corresponding probability tree, if not, creating a corresponding probability tree, and performing initialization, where the created probability tree is a binary tree, leaf nodes of the binary tree correspond to the available computing nodes, and each node except the root node has a probability value, where the probability value is a probability that a parent node of the node reaches the node.
In one embodiment, the cost value is determined by the following formula:
valuei＝probabilityi*RTTi
the cost function is determined by the following formula:
wherein, the mobilityiIs the probability that the ith matchable node is selected, RTTiIs the RTT delay, p, of the ith matchable node1,p2,…,pnCorresponding to n matchable nodes, D (p)1,p2,…,pn) K is an adjustment factor as a function of the running sum of the cost values for each matchable node.
According to a second aspect, an embodiment provides a method for deterministic forwarding of network resource requirements, comprising:
receiving a resource request packet, wherein the resource request packet comprises resource demand information;
searching a resource matching forwarding table according to the resource demand information, and obtaining a matchable node corresponding to the resource demand information, wherein each item of the resource matching forwarding table comprises resource demand information, the corresponding matchable node and attribute information of each matchable node;
calculating the cost value of each matched node according to the attribute information;
updating the probability tree corresponding to the matchable node by adopting an online learning method according to a predefined cost function, so that the probability of the leaf node corresponding to the matchable node with the minimum cost value and the non-leaf node passing from the root node to the leaf node is increased; the probability tree is a binary tree, and leaf nodes of the binary tree correspond to all matchable nodes;
selecting sub-nodes of the current node from the root node of the probability tree according to the probability of each sub-node of the current node until the selected node is a leaf node;
and forwarding the resource request packet to the matched node corresponding to the selected leaf node.
In one embodiment, the resource requirement information includes: resource requestor identification, task type of resource execution, input data location of the task, and resource demand.
In one embodiment, the attribute information includes resource tension, load balancing information, fairness information, and RTT delay.
In one embodiment, the method further comprises: and detecting whether the matchable node has a corresponding probability tree, if not, creating a corresponding probability tree, and initializing, wherein the created probability tree is a binary tree, leaf nodes of the binary tree correspond to the matchable nodes, each node except the root node has a probability value, and the probability value is the probability of a father node of the node reaching the node.
In one embodiment, an annealing algorithm is used in the process of updating the probability tree corresponding to the matchable node by adopting the online learning method.
In one embodiment, the cost value is determined by the following formula:
valuei＝probabilityi*RTTi
the cost function is determined by the following formula:
wherein, the mobilityiIs the probability that the ith matchable node is selected, RTTiIs the RTT delay, p, of the ith matchable node1,p2,…,pnCorresponding to n matchable nodes, D (p)1,p2,…,pn) K is an adjustment factor as a function of the running sum of the cost values for each matchable node.
According to a third aspect, an embodiment provides a computer-readable storage medium comprising a program executable by a processor to implement the above-described deterministic forwarding method for computational demands and deterministic forwarding method for network resource demands.
According to the network resource requirement, the deterministic forwarding method of the computational power requirement and the computer-readable storage medium of the embodiment, based on the model of the probability tree, the matching process of the resource requirement is converted into the process of weight adjustment of the probability tree, and the weight of the probability tree is adaptively adjusted according to the real-time change of the node resource condition by adopting the online learning method, so that the probability state of the probability tree approaches to the best and can adapt to the change of the network condition, finally, a node is selected according to the probability tree for forwarding the resource requirement, and the node with a better state is selected with a higher probability, thereby improving the efficiency of network resource requirement forwarding and the utilization benefit of network resources, realizing the load balance of the network nodes, and improving the performance of the network.
Drawings
FIG. 1 is a schematic diagram of a bipartite graph model according to which a deterministic forwarding method of network resource requirements is based;
fig. 2 is a flowchart illustrating a method for deterministic forwarding of network resource requirements according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of a probability tree involved in the deterministic forwarding method for network resource requirements according to an embodiment of the present invention;
fig. 4 is a schematic diagram illustrating adjustment of a probability tree in a deterministic forwarding method for network resource requirements according to an embodiment of the present invention;
FIG. 5 is a flow chart of a deterministic forwarding method for computing power requirements according to an embodiment of the present invention;
fig. 6 is a schematic diagram illustrating an adjustment of a probability tree in a deterministic forwarding method for computing power requirements according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the following detailed description and accompanying drawings. Wherein like elements in different embodiments are numbered with like associated elements. In the following description, numerous details are set forth in order to provide a better understanding of the present application. However, those skilled in the art will readily recognize that some of the features may be omitted or replaced with other elements, materials, methods in different instances. In some instances, certain operations related to the present application have not been shown or described in detail in order to avoid obscuring the core of the present application from excessive description, and it is not necessary for those skilled in the art to describe these operations in detail, so that they may be fully understood from the description in the specification and the general knowledge in the art.
Furthermore, the features, operations, or characteristics described in the specification may be combined in any suitable manner to form various embodiments. Also, the various steps or actions in the method descriptions may be transposed or transposed in order, as will be apparent to one of ordinary skill in the art. Thus, the various sequences in the specification and drawings are for the purpose of describing certain embodiments only and are not intended to imply a required sequence unless otherwise indicated where such sequence must be followed.
The numbering of the components as such, e.g., "first", "second", etc., is used herein only to distinguish the objects as described, and does not have any sequential or technical meaning. The term "connected" and "coupled" when used in this application, unless otherwise indicated, includes both direct and indirect connections (couplings).
The resource requirement information and the matchable node in the network can be regarded as belonging to two different sets respectively, and the two sets have no common elements, that is, the two sets are not intersected, so that a bipartite graph can be formed by taking the resource requirement information and the matchable node in the network as vertexes. Those skilled in the art will appreciate that the figures herein refer to a data structure. Referring to fig. 1, as shown in fig. 1, the resource requirement information r1, r2, r3 … … rn and the matchable nodes c1, c2, c3 … … cm are divided into two vertex sets of the bipartite graph, and finding a suitable matchable node for the resource requirement information is equivalent to connecting a suitable edge between the two vertex sets, so that the matching and forwarding of the resource requirement information in the network can be abstracted to the multiple matching problem of the bipartite graph, and the input parameters of the corresponding bipartite graph model include the following four parts:
(1) requirement matching topology G (R ═ C, E), where R ═ R { (R {)1,r2,…,rnRepresents n resource demand information in the network; c ═ C1,c2,…,cmRepresents m matchable nodes in the network; e { (r)n,cm),rn∈C,cmE C is a set of edges, representing node CmIs the resource demand information rnMay match a node.
(2) Weight matrix of bipartite graph
Wherein the element W in the W matrixn,mRepresenting matchable nodes cmAnd resource demand information rnThe matching profit is larger, and the larger the matching profit is, the more overall benefit can be brought to the network.
(3) Matchable node parameters
Wherein
Representing matchable nodes cmThe number of resource requirement information that has been matched.
(4) Resource demand information parameter
Wherein
Representing resource requirement information rnAnd to how many matchable nodes are allocated at the same time.
The output result of the model is: each resource requirement information rnAnd its correspondingly allocated matchable node cm。
The invention provides a deterministic forwarding method of network resource requirements by taking the theory as guidance, which can solve the problem of bipartite graph multiple matching and further forward resource requirement information to a proper matchable node. Referring to fig. 2, as shown in fig. 2, the method for deterministic forwarding of network resource requirement according to an embodiment of the present invention includes steps 101 to 108, which are described in detail below.
Step 101: a resource request packet is received. There should be two types of packets in the forwarding process of network resource requirement: a resource request packet and a resource matching result packet. The resource request packet is used for requesting resources from a network, and comprises resource demand information; the resource matching result packet is used for informing the resource requester of the result of the resource matching. In this embodiment, the resource request information and the resource matching result both adopt a hierarchical structure naming mode, that is, a similar URL naming mode. Specifically, the data format of the resource request information is as follows: resource requester identification/task type of resource execution/input data location/resource demand of task; the data format of the resource matching result is: the resource requester identifies/resource executes the task type/return result of resource matching.
Step 102: and searching a resource matching forwarding table according to the resource demand information in the resource request packet to obtain a matchable node corresponding to the resource demand information. The routing protocol and the service arrangement protocol of the network act together to generate a resource matching forwarding table. The format of the resource matching forwarding table is shown in table 1, where each entry in the table includes a resource requirement information and corresponding matchable nodes and attribute information of each matchable node, where A, B, C, D, E, F represents a node in the network, such as a router, an edge computation force node, and the like.
TABLE 1
Step 103: and detecting whether the matchable node has a corresponding probability tree. If the detection result is that there is no such corresponding probability tree, go to step 104; if the detection result is that there is such a corresponding probability tree, step 105 is performed.
Step 104: and creating a probability tree corresponding to the nodes which can be matched, and initializing. The created probability tree is a binary tree, leaf nodes of the binary tree correspond to the matched nodes, each node except the root node has a probability value, and the probability value is the probability of the parent node of the node reaching the node. The probability tree is created in a similar manner to the process of creating a huffman tree. Those skilled in the art will appreciate that the probability tree referred to in this disclosure is a data structure. Generally, as shown in fig. 3, the probability tree includes a root node, sub-nodes below the root node, sub-nodes at the lowest layer are called leaf nodes, and other sub-nodes are called non-leaf nodes. In one embodiment, the probability tree is initialized by equalizing the probabilities of the sub-nodes in the same layer. For example, in one embodiment, the probability tree may be initialized such that the probability of each child node in FIG. 3 is 1/2, i.e., for any child node, the probability of reaching that child node from its parent node is 1/2. It should be noted that the probability tree shown in fig. 3 is only an example of a probability tree, and the number of nodes shown in the figure is not used to limit the probability tree of the present invention to have only 10 sub-nodes.
Step 105: and calculating the cost value of each matched node. The cost value of a matchable node is related to its attribute information, including but not limited to resource tension, load balancing information, fairness information, RTT, Round-Trip Time (RTT). The attribute information can be dynamically updated in real time along with the change of the network state, so that the cost value reflects the quality of the current state of the matched node to a certain extent. In a specific implementation process, one or more specific attribute information may be selected according to actual requirements to calculate a cost value of a matchable node. For example, selecting the RTT delay to calculate the cost value, the cost value can be determined by the following formula:
valuei＝probabilityi*RTTi
wherein, the mobilityiIs the probability that the ith matchable node is selected, RTTiIs the RTT delay of the ith matchable node. The probability that a matchable node is selected refers to the probability that the leaf node corresponding to the matchable node is reached from the root node on the probability tree, for example, in fig. 3, if the probability values of the respective child nodes are 1/2, the probabilities that the matchable nodes corresponding to the four left-hand leaf nodes are selected are 1/8, and the rest of the nodes on the right are 1/8The probability that a matchable node corresponding to both leaf nodes is selected is 1/4.
Step 106: and updating the probability tree corresponding to the matchable node by adopting an online learning method according to a predefined cost function, so that the probability of the leaf node corresponding to the matchable node with the minimum cost value and the non-leaf node passing from the root node to the leaf node is increased. In order to adapt to the change of the network condition, so as to approach to the optimal state, the present embodiment defines the cost function of the probability tree, and updates the probability tree by using an online learning method according to the cost function of the probability tree, so as to increase the probability of the leaf node corresponding to the matchable node with the minimum cost value and the non-leaf node through which the leaf node reaches from the root node. The cost function of the probability tree is also related to the attribute information of the matchable nodes, including but not limited to resource tension, load balancing information, fairness information, RTT delay. The attribute information can be dynamically updated in real time along with the change of the network state. In a specific implementation process, one or more specific attribute information may be selected according to actual requirements to calculate a cost value of a matchable node. For example, selecting the RTT delay to define the cost function, the cost function can be determined by the following formula:
wherein, the mobilityiIs the probability that the ith matchable node is selected, RTTiIs the RTT delay, p, of the ith matchable node1,p2,…,pnCorresponding to n matchable nodes, D (p)1,p2,…,pn) K is an adjustment factor as a function of the running sum of the cost values for each matchable node.
As described above, the probability of each node refers to the probability that the parent node of each node reaches the node. Referring to fig. 4, it is assumed that the resource requirement information has 6 matches in the resource matching forwarding table, and forms the probability tree in fig. 4. Assuming the cost value of matchable node C is the smallest, the probability of its corresponding leaf node 3-3 is increased, i.e., the probability of reaching the leaf node 3-3 from the non-leaf node 2-2 is increased. At the same time, the probability of the non-leaf node that needs to be traversed to reach the leaf node 3-3 from the root node 0 increases, i.e., the probability of the non-leaf nodes 2-2 and 1-1 in the graph increases. The increased probability of non-leaf node 2-2 refers to an increased probability of reaching non-leaf node 2-2 from non-leaf node 1-1; the increased probability of non-leaf node 1-1 refers to an increased probability of reaching non-leaf node 1-1 from root node 0. Because the sum of the probabilities that any node reaches its respective child node is 1, when the probability of a leaf node 3-3 increases, the probability of a leaf node 3-4 decreases accordingly; when the probability of non-leaf node 2-2 increases, correspondingly, the probability of non-leaf node 2-1 decreases; as the probability of non-leaf node 1-1 increases, the probability of non-leaf node 2-3 decreases accordingly.
In one embodiment, an annealing algorithm is used during learning to avoid falling into local optima.
Step 107: and starting from the root node of the probability tree, selecting the sub-nodes of the current node according to the probability of each sub-node of the current node until the selected nodes are leaf nodes. For example, referring to FIG. 4, it is not assumed that in the latest probability tree, the probability of node 1-1 is 2/3 and the probability of node 2-3 is 1/3; the probability of node 2-1 is 1/4, and the probability of node 2-2 is 3/4; the probability of node 3-1 is 1/2, and the probability of node 3-2 is 1/2; the probability of node 3-3 is 3/4, and the probability of node 3-4 is 1/4; the probability of node 3-5 is 1/2, and the probability of node 3-6 is 1/2; accordingly, a probability value of 1/12 for the matchable node a, a probability value of 1/12 for the matchable node B, a probability value of 3/8 for the matchable node C, a probability value of 1/8 for the matchable node D, a probability value of 1/6 for the matchable node E, and a probability value of 1/6 for the matchable node F can be calculated. Starting from the root node of the probability tree, node 0 in the graph, there are 2/3 probability selection nodes 1-1 and 1/3 probability selection nodes 2-3; if node 1-1 is selected, then there is a probability of 1/4 to select node 2-1 and a probability of 3/4 to select node 2-2; if node 2-1 is selected, then there is a probability of 1/2 to select node 3-1 and a probability of 1/2 to select node 3-2; if node 3-2 is selected, selection is stopped because node 3-2 is now a leaf node.
Step 108: and forwarding the resource request packet to the matched node corresponding to the selected leaf node. As exemplified in step 107, the leaf node 3-2 is finally selected in fig. 4, and the resource request packet is forwarded to the matchable node B corresponding to the leaf node 3-2. The mode of selecting the matchable nodes according to the probability tree can ensure that the probability of the matchable nodes with larger probability values calculated according to the probability tree is higher when the matchable nodes are selected as the last actual forwarding nodes, so that the probability values of the matchable nodes can play a role in balancing network load. Through the learning process, the probability that the matchable node with the minimum cost value is selected is improved, so that the probability that the resource demand information is forwarded to the matchable node with a better state is increased, and the efficiency of network resource demand forwarding and the utilization benefit of network resources are improved.
The method for deterministically forwarding the network resource requirements provided by the invention can be applied to forwarding the requirements of computing resources, storage resources, bandwidth resources and the like in the network, and the application of the method for deterministically forwarding the network resource requirements provided by the invention in the computing service network is specifically described below by taking the computing requirement forwarding as an example.
Referring to fig. 5, as shown in fig. 5, a method for deterministic forwarding of computational power demand in one embodiment includes steps 201 to 208, which are described in detail below.
Step 201: and receiving an computing power request packet. In a computing power service network, two data packets exist in the forwarding process of computing power requirements: the computing power request packet and the computing power matching result packet. The computing power request packet is used for requesting computing power resources from the network, and comprises computing power demand information; the calculation power matching result packet is used for informing the calculation power requester of the result of calculation power matching. In this embodiment, the calculation power request information and the calculation power matching result both adopt a hierarchical structure naming mode, that is, a similar URL naming mode. Specifically, the data format of the computing power request information is as follows: calculating force requester identification/calculating task type/calculating input data position of task/calculating force demand; the data format of the computational power matching result is as follows: the computing requester identifies/computes task type/returned results of the computing power match.
As shown in table 2, the calculation power requirement information of the face recognition calculation task in the university of beijing institute of information and engineering a116 room can be expressed in the following format: PKU/SECE/A116/faceRecognition/DataInput1/32_2TS/2G _ Memory. Wherein PKU/SECE/A116 represents the Beijing university institute of information and engineering A116 room; the FaceRecognition represents a face recognition computing task; DataInput1 represents the location of input data for the face recognition computation task; 32_2TS represents the computing power required by the face recognition computing task, namely the 32-bit floating point computing power which needs to be operated for 2 trillion times per second; 2G _ Memory represents that the computing task requires 2G Memory. Similarly, the calculation power requirement information of the speech recognition calculation task in the beijing university dormitory No. 3 building 1413 room can be expressed in the following format: PKU/Dormitry _3_1413/SpeechRecognitio n/DataInput2/32_1TS/1G _ Memory.
TABLE 2
As shown in table 3, the results of the computation power matching for the face recognition computation task in the university of beijing institute of information and engineering a116 room can be expressed in the following format: PKU/SECE/A116/faceRecognation/Result, wherein PKU/SECE/A116 represents the Beijing university institute of information and engineering A116 room; the FaceRecognition represents a face recognition computing task; result is the return Result of the computational power match.
TABLE 3
Step 202: and searching a calculation power matching forwarding table according to the calculation power demand information in the calculation power request packet to obtain an available calculation node corresponding to the calculation power demand information. The routing protocol and the service arrangement protocol of the computing power service network act together to generate a computing power matching forwarding table CIB (computing Information base). The format of the computation power matching forwarding table is shown in table 4, where each entry in the table includes a computation power requirement information and corresponding available computation nodes and attribute information of each available computation node, where A, B, C, D, E, F represents computation nodes in the computation power service network, such as routers, edge computation power nodes, etc.
TABLE 4
Step 203: and detecting whether the available computing nodes have a corresponding probability tree. If the detection result is that there is no such corresponding probability tree, go to step 204; if the detection result is that there is such a corresponding probability tree, step 205 is performed.
Step 204: and creating a probability tree corresponding to the available computing nodes and initializing. The created probability tree is a binary tree, leaf nodes of the binary tree correspond to all available computing nodes, each node except a root node has a probability value, and the probability value is the probability of a father node of the node reaching the node. The probability tree is created in a similar manner to the process of creating a huffman tree. Those skilled in the art will appreciate that the probability tree referred to in this disclosure is a data structure. Generally, as shown in fig. 3, the probability tree includes a root node, sub-nodes below the root node, sub-nodes at the lowest layer are called leaf nodes, and other sub-nodes are called non-leaf nodes. In one embodiment, the probability tree is initialized by equalizing the probabilities of the sub-nodes in the same layer. For example, in one embodiment, the probability tree may be initialized such that the probability of each child node in FIG. 3 is 1/2, i.e., for any child node, the probability of reaching that child node from its parent node is 1/2. It should be noted that the probability tree shown in fig. 3 is only an example of a probability tree, and the number of nodes shown in the figure is not used to limit the probability tree of the present invention to have only 10 sub-nodes.
Step 205: the cost value of each available compute node is computed. The cost value of an available compute node is related to its attribute information, including but not limited to computational resource tension, load balancing information, fairness information, RTT Time (Round-Trip Time). The attribute information can be dynamically updated in real time along with the change of the network state, so that the cost value reflects the current state of the available computing node to a certain extent. In a specific implementation process, one or more specific attribute information can be selected according to actual requirements to calculate the cost value of the available computing nodes. For example, selecting the RTT delay to calculate the cost value, the cost value can be determined by the following formula:
valuei＝probabilityi*RTTi
wherein, the mobilityiIs the probability that the ith available computing node is selected, RTTiIs the RTT delay of the ith available computing node. The probability that an available computation node is selected refers to the probability that a leaf node corresponding to the available computation node is reached from the root node on the probability tree, for example, in fig. 3, if the probability values of the respective sub-nodes are 1/2, the probabilities that the available computation nodes corresponding to the four left leaf nodes are selected are 1/8, and the probabilities that the available computation nodes corresponding to the remaining two right leaf nodes are selected are 1/4.
Step 206: and updating the probability tree corresponding to the available calculation node by adopting an online learning method according to a predefined cost function, so that the probability of the leaf node corresponding to the available calculation node with the minimum cost value and the non-leaf node passing through the leaf node from the root node is increased. In order to adapt to the change of the network condition, so as to approach to the optimal state, the present embodiment defines the cost function of the probability tree, and updates the probability tree by using an online learning method according to the cost function of the probability tree, so that the probabilities of the leaf node corresponding to the available computation node with the minimum cost value and the non-leaf node through which the leaf node reaches from the root node are increased. The cost function of the probability tree is also related to the attribute information of the available compute nodes, including but not limited to resource tension, load balancing information, fairness information, RTT delay. The attribute information can be dynamically updated in real time along with the change of the network state. In a specific implementation process, one or more specific attribute information can be selected according to actual requirements to calculate the cost value of the available computing nodes. For example, selecting the RTT delay to define the cost function, the cost function can be determined by the following formula:
wherein, the mobilityiIs the probability that the ith available computing node is selected, RTTiIs the RTT delay, p, of the ith available computing node1,p2,…,pnCorresponding to n available computing nodes, D (p)1,p2,…,pn) K is an adjustment factor for the running sum function of the cost value of each available compute node.
As described above, the probability of each node refers to the probability that the parent node of each node reaches the node. Referring to fig. 6, it is assumed that the computation power requirement information has 6 available computation nodes in the computation power matching forwarding table, and forms the probability tree in fig. 6. Assuming the cost value of the available compute node C is the smallest, the probability of its corresponding leaf node 3-3 is increased, i.e., the probability of reaching the leaf node 3-3 from the non-leaf node 2-2 is increased. At the same time, the probability of the non-leaf node that needs to be traversed to reach the leaf node 3-3 from the root node 0 increases, i.e., the probability of the non-leaf nodes 2-2 and 1-1 in the graph increases. The increased probability of non-leaf node 2-2 refers to an increased probability of reaching non-leaf node 2-2 from non-leaf node 1-1; the increased probability of non-leaf node 1-1 refers to an increased probability of reaching non-leaf node 1-1 from root node 0. Because the sum of the probabilities that any node reaches its respective child node is 1, when the probability of a leaf node 3-3 increases, the probability of a leaf node 3-4 decreases accordingly; when the probability of non-leaf node 2-2 increases, correspondingly, the probability of non-leaf node 2-1 decreases; as the probability of non-leaf node 1-1 increases, the probability of non-leaf node 2-3 decreases accordingly.
In one embodiment, an annealing algorithm is used during learning to avoid falling into local optima.
Step 207: and starting from the root node of the probability tree, selecting the sub-nodes of the current node according to the probability of each sub-node of the current node until the selected nodes are leaf nodes. For example, referring to FIG. 6, it is not assumed that in the latest probability tree, the probability of node 1-1 is 2/3 and the probability of node 2-3 is 1/3; the probability of node 2-1 is 1/4, and the probability of node 2-2 is 3/4; the probability of node 3-1 is 1/2, and the probability of node 3-2 is 1/2; the probability of node 3-3 is 3/4, and the probability of node 3-4 is 1/4; the probability of node 3-5 is 1/2, and the probability of node 3-6 is 1/2; accordingly, a probability value of 1/12 for available compute node a, a probability value of 1/12 for available compute node B, a probability value of 3/8 for available compute node C, a probability value of 1/8 for available compute node D, a probability value of 1/6 for available compute node E, and a probability value of 1/6 for available compute node F may be calculated. Starting from the root node of the probability tree, node 0 in the graph, there are 2/3 probability selection nodes 1-1 and 1/3 probability selection nodes 2-3; if node 1-1 is selected, then there is a probability of 1/4 to select node 2-1 and a probability of 3/4 to select node 2-2; if node 2-1 is selected, then there is a probability of 1/2 to select node 3-1 and a probability of 1/2 to select node 3-2; if node 3-2 is selected, selection is stopped because node 3-2 is now a leaf node.
Step 208: and forwarding the calculation force request packet to the available calculation node corresponding to the selected leaf node. As exemplified in step 207, the leaf node 3-2 is eventually selected in fig. 6, and the computation request packet is forwarded to the available compute node B corresponding to the leaf node 3-2. The method for selecting the available computing nodes according to the probability tree can make the probability of the available computing nodes with higher probability values obtained by computing according to the probability tree higher, the probability of the available computing nodes selected as the last actual forwarding nodes is higher, and therefore the probability values of the available computing nodes can play a role in balancing network load. Through the learning process, the probability that the available computing node with the minimum cost value is selected is improved, so that the probability that the computing power demand information is forwarded to the available computing node with a better state is increased, and the efficiency of computing power demand forwarding and the utilization benefit of computing power resources are improved.
The invention theoretically models the forwarding of the resource demand information in the network into a bipartite graph multiple matching problem, semantically abstracts the resource demand information and the resource matching result, and provides a network resource demand and calculation power demand deterministic forwarding method based on probability tree and online learning. The method of the invention converts the matching process of the resource requirement into the process of weight adjustment of the probability tree based on the model of the probability tree, and adjusts the weight of the probability tree according to the real-time change of the node resource status in a self-adaptive manner by adopting the method of online learning, so that the probability state of the probability tree approaches to the best and can adapt to the change of the network status, and finally, a node is selected according to the probability tree to forward the resource requirement, and the probability of selecting the node with better state is higher, thereby improving the efficiency of forwarding the network resource requirement and the utilization benefit of the network resource, realizing the load balance of the network node, and improving the performance of the network.
Reference is made herein to various exemplary embodiments. However, those skilled in the art will recognize that changes and modifications may be made to the exemplary embodiments without departing from the scope hereof. For example, the various operational steps, as well as the components used to perform the operational steps, may be implemented in differing ways depending upon the particular application or consideration of any number of cost functions associated with operation of the system (e.g., one or more steps may be deleted, modified or incorporated into other steps).
In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. Additionally, as will be appreciated by one skilled in the art, the principles herein may be reflected in a computer program product on a computer readable storage medium, which is pre-loaded with computer readable program code. Any tangible, non-transitory computer-readable storage medium may be used, including magnetic storage devices (hard disks, floppy disks, etc.), optical storage devices (CD-to-ROM, DVD, Blu-Ray discs, etc.), flash memory, and/or the like. These computer program instructions may be loaded onto a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions which execute on the computer or other programmable data processing apparatus create means for implementing the functions specified. These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including means for implementing the function specified. The computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified.
While the principles herein have been illustrated in various embodiments, many modifications of structure, arrangement, proportions, elements, materials, and components particularly adapted to specific environments and operative requirements may be employed without departing from the principles and scope of the present disclosure. The above modifications and other changes or modifications are intended to be included within the scope of this document.
The foregoing detailed description has been described with reference to various embodiments. However, one skilled in the art will recognize that various modifications and changes may be made without departing from the scope of the present disclosure. Accordingly, the disclosure is to be considered in an illustrative and not a restrictive sense, and all such modifications are intended to be included within the scope thereof. Also, advantages, other advantages, and solutions to problems have been described above with regard to various embodiments. However, the benefits, advantages, solutions to problems, and any element(s) that may cause any element(s) to occur or become more pronounced are not to be construed as a critical, required, or essential feature or element of any or all the claims. As used herein, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, system, article, or apparatus. Furthermore, the term "coupled," and any other variation thereof, as used herein, refers to a physical connection, an electrical connection, a magnetic connection, an optical connection, a communicative connection, a functional connection, and/or any other connection.
Those skilled in the art will recognize that many changes may be made to the details of the above-described embodiments without departing from the underlying principles of the invention. Accordingly, the scope of the invention should be determined only by the claims.
Claims (10)
1. A method for deterministic forwarding of computational power demands, comprising:
receiving a calculation force request packet, wherein the calculation force request packet comprises calculation force demand information; the calculation force demand information comprises calculation force requester identification, calculation task types, input data positions of calculation tasks and calculation force demand quantity;
searching a calculation power matching forwarding table according to the calculation power demand information, and obtaining available computing nodes corresponding to the calculation power demand information, wherein each item of the calculation power matching forwarding table comprises calculation power demand information, the corresponding available computing nodes and attribute information of each available computing node; the attribute information comprises computing resource tension, load balancing information, fairness information and RTT time delay;
calculating the cost value of each available computing node according to the attribute information;
updating the probability tree corresponding to the available computing node by adopting an online learning method according to a predefined cost function, so that the probability of a leaf node corresponding to the available computing node with the minimum cost value and a non-leaf node passing through the leaf node from a root node is increased; the probability tree is a binary tree, and leaf nodes of the binary tree correspond to all available computing nodes;
selecting sub-nodes of the current node from the root node of the probability tree according to the probability of each sub-node of the current node until the selected node is a leaf node;
and forwarding the computing power request packet to the available computing node corresponding to the selected leaf node.
2. The method of claim 1, further comprising:
and detecting whether the available computing node has a corresponding probability tree, if not, creating a corresponding probability tree, and initializing, wherein the created probability tree is a binary tree, leaf nodes of the binary tree correspond to the available computing nodes, each node except the root node has a probability value, and the probability value is the probability of a father node of the node reaching the node.
3. The method of claim 1, wherein the cost value is determined by the formula:
valuei＝probabilityi*RTTi
the cost function is determined by the following formula:
wherein, the mobilityiIs the probability that the ith matchable node is selected, RTTiIs the RTT delay, p, of the ith matchable node1,p2,…,pnCorresponding to n matchable nodes, D (p)1,p2,…,pn) K is an adjustment factor as a function of the running sum of the cost values for each matchable node.
4. A method for deterministic forwarding of network resource requirements, comprising:
receiving a resource request packet, wherein the resource request packet comprises resource demand information;
searching a resource matching forwarding table according to the resource demand information, and obtaining a matchable node corresponding to the resource demand information, wherein each item of the resource matching forwarding table comprises resource demand information, the corresponding matchable node and attribute information of each matchable node;
calculating the cost value of each matched node according to the attribute information;
updating the probability tree corresponding to the matchable node by adopting an online learning method according to a predefined cost function, so that the probability of the leaf node corresponding to the matchable node with the minimum cost value and the non-leaf node passing from the root node to the leaf node is increased; the probability tree is a binary tree, and leaf nodes of the binary tree correspond to all matchable nodes;
selecting sub-nodes of the current node from the root node of the probability tree according to the probability of each sub-node of the current node until the selected node is a leaf node;
and forwarding the resource request packet to the matched node corresponding to the selected leaf node.
5. The method of claim 4, wherein the resource demand information comprises: resource requestor identification, task type of resource execution, input data location of the task, and resource demand.
6. The method of claim 4, wherein the attribute information comprises resource tension, load balancing information, fairness information, and RTT delay.
7. The method of claim 4, further comprising:
and detecting whether the matchable node has a corresponding probability tree, if not, creating a corresponding probability tree, and initializing, wherein the created probability tree is a binary tree, leaf nodes of the binary tree correspond to the matchable nodes, each node except the root node has a probability value, and the probability value is the probability of a father node of the node reaching the node.
8. The method of claim 4, wherein an annealing algorithm is used in the process of updating the probability tree corresponding to the matchable node by adopting the online learning method.
9. The method of claim 6, wherein the cost value is determined by the formula:
valuei＝probabilityi*RTTi
the cost function is determined by the following formula:
wherein, the mobilityiIs the probability that the ith matchable node is selected, RTTiIs the RTT delay, p, of the ith matchable node1,p2,…,pnCorresponding to n matchable nodes, D (p)1,p2,…,pn) K is an adjustment factor as a function of the running sum of the cost values for each matchable node.
10. A computer-readable storage medium, characterized by comprising a program executable by a processor to implement the method of any one of claims 1 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010773478.3A CN112003907A (en) | 2020-08-04 | 2020-08-04 | Deterministic forwarding method for network resource demand and computational power demand thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010773478.3A CN112003907A (en) | 2020-08-04 | 2020-08-04 | Deterministic forwarding method for network resource demand and computational power demand thereof |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112003907A true CN112003907A (en) | 2020-11-27 |
Family
ID=73463623
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010773478.3A Pending CN112003907A (en) | 2020-08-04 | 2020-08-04 | Deterministic forwarding method for network resource demand and computational power demand thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112003907A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115086230A (en) * | 2022-06-15 | 2022-09-20 | 中国联合网络通信集团有限公司 | Method, device, equipment and storage medium for determining routing of computing network |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106534360A (en) * | 2016-12-13 | 2017-03-22 | 重庆大学 | Content centric network routing method based on software definition |
| US20180309666A1 (en) * | 2016-08-18 | 2018-10-25 | Peking University Shenzhen Graduate School | Method and apparatus for forwarding an interest packet based on a probability tree in an information network |
| CN108924192A (en) * | 2018-06-15 | 2018-11-30 | 中国人民解放军国防科技大学 | Optimal task scheduling method and system based on pseudo tree structure under data center network |
| CN109413117A (en) * | 2017-08-15 | 2019-03-01 | 中兴通讯股份有限公司 | Distributed data calculation method, device, server and computer storage medium |
| CN109818786A (en) * | 2019-01-20 | 2019-05-28 | 北京工业大学 | A kind of cloud data center applies the more optimal choosing methods in combination of resources path of appreciable distribution |
| CN110290069A (en) * | 2019-06-18 | 2019-09-27 | 西安交通大学 | A kind of reliable multicast method for routing under SDN scene |
-
2020
- 2020-08-04 CN CN202010773478.3A patent/CN112003907A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180309666A1 (en) * | 2016-08-18 | 2018-10-25 | Peking University Shenzhen Graduate School | Method and apparatus for forwarding an interest packet based on a probability tree in an information network |
| CN106534360A (en) * | 2016-12-13 | 2017-03-22 | 重庆大学 | Content centric network routing method based on software definition |
| CN109413117A (en) * | 2017-08-15 | 2019-03-01 | 中兴通讯股份有限公司 | Distributed data calculation method, device, server and computer storage medium |
| CN108924192A (en) * | 2018-06-15 | 2018-11-30 | 中国人民解放军国防科技大学 | Optimal task scheduling method and system based on pseudo tree structure under data center network |
| CN109818786A (en) * | 2019-01-20 | 2019-05-28 | 北京工业大学 | A kind of cloud data center applies the more optimal choosing methods in combination of resources path of appreciable distribution |
| CN110290069A (en) * | 2019-06-18 | 2019-09-27 | 西安交通大学 | A kind of reliable multicast method for routing under SDN scene |
Non-Patent Citations (1)
| Title |
|---|
| 黄少芬等: "基于贝叶斯概率树的机会社会网络转发节点选择方法", 《安徽师范大学学报(自然科学版)》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115086230A (en) * | 2022-06-15 | 2022-09-20 | 中国联合网络通信集团有限公司 | Method, device, equipment and storage medium for determining routing of computing network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Brest et al. | Performance comparison of self-adaptive and adaptive differential evolution algorithms | |
| US10491464B2 (en) | Network topology assisted device provisioning | |
| EP2835938B1 (en) | Message publishing and subscribing method and apparatus | |
| JP4736713B2 (en) | Systems and methods to support the selection of project members | |
| Peng et al. | A fast algorithm to find all-pairs shortest paths in complex networks | |
| CN112583716A (en) | Path selection method and device based on data transmission of Internet of things | |
| WO2024036694A1 (en) | Control instruction sending method, control apparatus, storage medium and electronic apparatus | |
| CN112003907A (en) | Deterministic forwarding method for network resource demand and computational power demand thereof | |
| CN115543556A (en) | Adaptive symbolic regression method based on multitask genetic programming algorithm | |
| US8756093B2 (en) | Method of monitoring a combined workflow with rejection determination function, device and recording medium therefor | |
| Gandhi et al. | Performance comparison of parallel graph coloring algorithms on bsp model using hadoop | |
| Hui et al. | Top-k query framework in wireless sensor networks for smart grid | |
| JP6303032B2 (en) | Method and apparatus for generating instances of technical indicators | |
| CN114422582B (en) | Dynamic service combination method and device for technological resources | |
| CN110300168A (en) | A kind of concurrent tasks optimization method based on multiple edge server collaboration | |
| Tun et al. | Optimizing resource discovery technique in the P2P grid systems | |
| US10445080B2 (en) | Methods for adaptive placement of applications and devices thereof | |
| CN115034615A (en) | Method for improving feature selection efficiency in genetic programming scheduling rule for job shop scheduling | |
| Nasir et al. | Fast trust computation in online social networks | |
| Huang et al. | A novel Bi-Ant colony optimization algorithm for solving multi-objective service selection problem | |
| CN108932262A (en) | A kind of song recommendations method and device | |
| CN110020087B (en) | Distributed PageRank acceleration method based on similarity estimation | |
| Zhang et al. | A policy-driven service composition method for adaptation in pervasive computing environment | |
| CN116760762B (en) | Decentralised ad hoc network method and device | |
| Yingqiu et al. | Cloud task scheduling based on chaotic particle swarm optimization algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20201127 |