WO2015021869A1

WO2015021869A1 - Method and device for selecting optimal network maximum flow algorithm

Info

Publication number: WO2015021869A1
Application number: PCT/CN2014/083511
Authority: WO
Inventors: 王蕾; 崔慧敏; 吕方; 冯晓兵
Original assignee: 华为技术有限公司
Priority date: 2013-08-14
Filing date: 2014-08-01
Publication date: 2015-02-19
Also published as: CN104376366A; CN104376366B

Abstract

Embodiments of the present invention provide a method and a device for selecting optimal network maximum flow algorithm, and relates to the field of network maximum flows, which can determine an optimal network maximum flow algorithm according to different network flow graphs. The method comprises: an algorithm selecting device acquiring a network flow graph, and obtaining a first residual network according to the network flow graph; acquiring an operation algorithm set, the operation algorithm set comprising at least two algorithms; performing preflow push concurrently in the first residual network by using at least two algorithms, to obtain a second residual network, determining, in the second residual network, a quantity of pivotal edges corresponding to the at least two algorithms, and determining that an algorithm corresponding to the maximum value of the quantity of pivotal edges, as an optimal network maximum flow algorithm of the network flow graph. The embodiments of the present invention are used for selecting a network maximum flow algorithm.

Description

Method and device for selecting optimal network maximum flow algorithm

TECHNICAL FIELD The present invention relates to the field of network maximum flow, and in particular, to a method and device for selecting an optimal network maximum stream algorithm. BACKGROUND OF THE INVENTION Acquiring the maximum flow problem of a network is a classic problem in graph theory and combinatorial optimization, and it has a wide application background.

In the prior art, the pre-flow propulsion method is a commonly used method for obtaining a maximum flow of a network, and the method includes multiple algorithms, which can be summarized as: obtaining a residual network according to the acquired network flow graph, from the residual network Obtaining an active node, and in the residual network, the surplus stream corresponding to the active node is advanced to the sink point through the adjacent node, and after determining that the adjacent node is a new active node, continuing in the The residual network pushes the new active node to the sink point through the adjacent node until the active node does not exist in the residual network, and then determines that the traffic of the sink point is the maximum flow of the network.

It can be seen from the above that in the process of obtaining the maximum flow of the network by the pre-flow advancement method, only the network flow graph is pre-flowed by an algorithm, that is, regardless of the type of the network flow graph acquired ( For example, it can be an unbalanced bipartite graph), which is calculated by an algorithm. However, for different types of network flow graphs, the calculation of the maximum flow of the algorithm is not necessarily optimal, because the network flow graph is not adapted. The algorithm has many redundant operations, so the algorithm performance of the algorithm on the network flow graph (such as operation time and computational efficiency) cannot be guaranteed, thereby reducing the efficiency of the network maximum stream calculation. SUMMARY OF THE INVENTION Embodiments of the present invention provide a method and a device for selecting an optimal network maximum flow algorithm, which are capable of determining an optimal network maximum flow algorithm according to different network flow graphs.

In order to achieve the above object, embodiments of the present invention use the following technical solutions:

In a first aspect, a method for selecting an optimal network maximum flow algorithm is provided, including: Obtaining a network flow graph, and obtaining a first residual network according to the network flow graph;

Obtaining a set of operation algorithms, where the set of operation algorithms includes at least two algorithms; performing pre-flow advancement in parallel by the at least two algorithms in the first residual network to obtain a second residual network, and Determining, in the second residual network, the number of key edges corresponding to the at least two algorithms;

The algorithm corresponding to determining the maximum value of the number of key edges is an optimal network maximum flow algorithm of the network flow graph.

In a first possible implementation manner of the first aspect, after the obtaining the first residual network according to the network flow graph, the method further includes:

Obtaining an active node according to the first residual network;

The acquisition operation algorithm set includes:

Acquiring an algorithm and a data structure corresponding to the active node according to the algorithm template, where the algorithm template includes a data structure configured by the user to manage the active node, and an algorithm configured by the user corresponding to the active node, and the algorithm The data structure of the network flow graph;

The set of operational algorithms is obtained in accordance with the algorithm and data structure.

In conjunction with the first possible implementation, in a second possible implementation, the method further includes:

Fine-grained locking of the data structure of the network flow graph.

In combination with the first possible implementation manner or the second possible implementation manner, in a third possible implementation manner, the pre-flow advancement is performed in parallel by the at least two algorithms in the first residual network. Includes:

Obtaining a corresponding active node from the data structure of the management active node according to the at least two algorithms;

Performing, according to the at least two algorithms, the surplus stream of the corresponding active node in parallel to the sink point by using an adjacent node whose distance identifier is lower than itself, wherein the surplus stream includes the input of the active node The positive difference between the flow and the outgoing flow.

With reference to the third possible implementation, in a fourth possible implementation, after the surplus flow of the corresponding active node is advanced to the sink point by using the adjacent node, the method further includes:

If the neighboring node has the surplus flow, determining that the adjacent node is new active a node, and placing the new active node in the data structure of the management active node. In combination with the fourth possible implementation manner, in the fifth possible implementation manner, before the obtaining the second residual network, the method further includes:

Obtaining the number of edges that are re-identified when the pre-flow advancement is performed in parallel by the at least two algorithms;

Obtaining an update parameter according to the algorithm template;

And after determining that the update parameter and the number of re-identified edges satisfy a preset condition, updating a distance identifier of all nodes to the sink point to obtain a new distance identifier of all the nodes to the sink point;

The obtaining the second residual network includes:

The second residual network is obtained according to the new distance identifier of all the nodes to the sink. In combination with the fourth possible implementation manner or the fifth possible implementation manner, in a sixth possible implementation manner, when determining that the data structure of the management active node does not have an active node, determining the sink point The traffic is the maximum flow of the network.

With reference to any one of the possible implementations of the first aspect to the sixth possible implementation, in a seventh possible implementation, the at least two algorithms are used in the first residual network Before the pre-flow advancement is performed in parallel, the method further includes:

Get the total number of threads;

Configuring a number of threads corresponding to the at least two algorithms according to the total number of threads, where the number of threads includes a ratio of the total number of threads to an algorithm number of the at least two algorithms; The pre-flow advancement in parallel in the residual network by the at least two algorithms includes:

Pre-flow advancement is performed in parallel by the at least two algorithms according to the corresponding number of threads in the first residual network.

With reference to the seventh possible implementation manner, in the eighth possible implementation manner, after determining that the algorithm corresponding to the maximum value of the number of the critical edges is the optimal network maximum flow algorithm, the method further includes:

The number of threads is reconfigured for the at least two algorithms based on the number of key edges.

In a second aspect, an algorithm selection device is provided, including:

An obtaining unit, configured to acquire a network flow graph, and obtain a first residual amount according to the network flow graph Networking, and acquiring a set of operational algorithms, where the set of operational algorithms includes at least two algorithms;

a processing unit, configured to perform pre-flow advancement in parallel by the at least two algorithms in the first residual network, to obtain a second residual network, and determine a key corresponding to the at least two algorithms after pre-flow advancement An algorithm corresponding to the number of edges and determining a maximum value of the number of key edges in the second residual network is an optimal network maximum flow algorithm of the network flow graph.

In a first possible implementation manner of the second aspect, the acquiring unit is further configured to: after obtaining the first residual network according to the network flow graph, acquiring an active node according to the first residual network;

The acquiring unit is specifically configured to: acquire an algorithm and a data structure corresponding to the active node according to an algorithm template, and obtain the operation algorithm set according to the algorithm and the data structure, where the algorithm template includes user configured management The data structure of the active node and the user-configured algorithm corresponding to the active node and the data structure of the network flow graph.

In conjunction with the first possible implementation, in a second possible implementation, the processing unit is further configured to add a fine-grained lock to the data structure of the network flow graph.

In combination with the first possible implementation manner or the second possible implementation manner, in a third possible implementation manner, the processing unit is specifically configured to perform an active active node according to the at least two algorithms, Acquiring a corresponding active node in the data structure, and advancing the surplus stream of the corresponding active node in parallel with the neighboring node whose distance identifier is lower than itself according to the at least two algorithms, wherein The surplus stream includes a positive difference between the inflow and the outflow of the active node.

In conjunction with the third possible implementation, in a fourth possible implementation, the processing unit is further configured to: perform a surplus flow of the corresponding active node to the sink point by using the neighboring node After advancing, if the neighboring node has the surplus stream, determining that the adjacent node is a new active node, and placing the new active node into the data of the management active node In the structure.

With reference to the fourth possible implementation, in a fifth possible implementation, the processing unit is further configured to: when the pre-flow advancement is performed in parallel by the at least two algorithms, obtain the number of edges that are re-identified, Obtaining an update parameter according to the algorithm template, and when determining that the update parameter and the number of re-identified edges satisfy a preset condition, updating a distance identifier of all nodes to the sink point to obtain all the nodes to The new distance identifier of the meeting point, and according to all the knots mentioned The new distance identification point to the sink point gets the second residual network.

In combination with the fourth possible implementation manner or the fifth possible implementation manner, in a sixth possible implementation manner, the processing unit is further configured to: when determining that the data structure of the management active node does not have an active node When the point is reached, it is determined that the traffic of the sink is the maximum flow of the network.

With reference to any one of the possible implementations of the second aspect to the sixth possible implementation, in a seventh possible implementation, the acquiring unit is further configured to: pass in the first residual network Obtaining a total number of threads before the at least two algorithms perform the pre-flow advancement in parallel; the processing unit is further configured to: configure a number of threads corresponding to the at least two algorithms, and pass the network in the first residual network The at least two algorithms perform pre-flow advancement according to the corresponding number of threads, wherein the number of threads includes a ratio of the total number of threads to the number of algorithms of the at least two algorithms.

With reference to the seventh possible implementation manner, in an eighth possible implementation, the processing unit is further configured to: after determining that the algorithm corresponding to the maximum value of the number of the critical edges is the optimal network maximum flow algorithm, The number of threads is reconfigured for the at least two algorithms based on the number of key edges.

Using the above scheme, the network flow graph is pre-flowed by operating a plurality of algorithms in the algorithm set, and after the pre-flow advancement, the algorithm with the largest number of key edges corresponding to each algorithm is determined as the optimal network maximum flow algorithm. In the process of obtaining the maximum flow of the network, the adaptive network flow graph can be dynamically adjusted to obtain an optimal performance algorithm, thereby improving the efficiency of the network maximum flow calculation. BRIEF DESCRIPTION OF THE DRAWINGS In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly described below. It is obvious that the drawings in the following description are only some implementations of the present invention. For example, other drawings may be obtained from those of ordinary skill in the art in light of the inventive work.

FIG. 1 is a schematic flowchart of a method for selecting an optimal network maximum flow algorithm according to an embodiment of the present invention;

FIG. 2 is a schematic flowchart diagram of another method for selecting an optimal network maximum flow algorithm according to an embodiment of the present invention; FIG. 3 is a schematic structural diagram of an algorithm selection device according to an embodiment of the present invention; FIG. 4 is a schematic structural diagram of another algorithm selection device according to an embodiment of the present invention. The technical solutions in the embodiments of the present invention are clearly and completely described in the following with reference to the accompanying drawings in the embodiments of the present invention. It is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. example. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts are within the scope of the present invention.

The embodiment of the invention provides a method for selecting an optimal network maximum flow algorithm. As shown in FIG. 1 , the execution subject of the method is an algorithm selection device, including:

S101. The algorithm selection device acquires a network flow graph, and obtains a first residual network according to the network flow graph.

The residual network is obtained by transforming the network flow graph.

S102: The algorithm selection device acquires a set of operation algorithms.

The set of operational algorithms includes at least two algorithms.

Specifically, the algorithm selecting device acquires an active node according to the first residual network, where the active node is a node whose inbound traffic is greater than the outgoing traffic, and the algorithm selecting device acquires an algorithm and data corresponding to the active node according to the algorithm template. Structure, and the set of operational algorithms is obtained according to the algorithm and data structure.

The algorithm template includes a data structure configured by the user to manage the active node, an algorithm configured by the user corresponding to the active node, and a data structure of the network flow graph.

For example, the algorithm template includes a main control part, providing a framework of the algorithm; the algorithm part, the user may configure at least two algorithms; the node management part, including the acquired active node and the user-configured corresponding management active node data Structure (such as bucke t, F IFO-queue and s tack); update event, used to update the distance identifier of each node in the first residual network to the sink; update parameters, configured by the user; Network flow graph data structure for storing residual networks.

The algorithm part, the node management part and the update parameter can be configured by the user. For example, the above algorithm framework can be: Whi le (act i veV) {

i = workse t. remove ();

Ver tex. compute (i);

Where activeV indicates the number of active nodes in the residual network, while (activeV) indicates that the number of active nodes in the residual network is not zero, the algorithm is executed, i = workset. remove 0 indicates that the active node is In the node management part, the workset represents the data structure of the corresponding active management node configured by the user; vertex, compute (i) indicates that the surplus stream of the active node is operated by the algorithm corresponding to the active node.

It should be noted that the algorithm selection device may obtain an algorithm and a data structure configured by the user according to the foregoing algorithm template, and obtain the operation algorithm set according to the algorithm and the data structure, where the operation algorithm set may be:

Workse t:: bucket (i->d);

Vertex, computel (node* i) { ... workset. push ( j); } vertex. compute2 (node* i) { ... }

GLOB_UPDT-FREQ = 0.5;

Where workset:: bucket (i->d) represents the data structure of the user-managed active node; vertex, computel (node* i) { ... workset. push ( j); } indicates the active node and the corresponding algorithm , GL0B_UPDT_FREQ = 0.5 indicates the user-configured update parameters. Parallel operations of multiple algorithms can be implemented by the above set of operational algorithms.

Further, the algorithm selection device may fine-grain the data structure of the network flow graph to prevent data competition, for example, preventing other algorithms other than the active node corresponding algorithm from reading and writing the active node.

S103. The algorithm selection device performs pre-flow advancement in parallel in the first residual network by using the at least two algorithms to obtain a second residual network, and determines, in the second residual network, the at least two algorithms. The number of key edges.

Here, the parallelism refers to that another algorithm can be executed at the same time in the process of executing one of the algorithms, and is not limited to starting at the same time.

Specifically, the algorithm selects the number of active nodes from the management according to the at least two algorithms According to the structure, the corresponding active node is obtained, and the surplus stream of the corresponding active node is pushed in parallel to the sink point by the adjacent node whose distance identifier is lower than the self according to the at least two algorithms.

The surplus flow includes a positive difference between the inflow and the outflow of the active node.

Optionally, if the neighboring node has the surplus stream, the algorithm selecting device determines that the neighboring node is a new active node, and places the new active node into the data structure of the managing active node. in.

Further, when the pre-streaming advancement is performed in parallel by the at least two algorithms, the algorithm selecting device acquires the number of re-identified edges, and obtains an update parameter according to the algorithm template, and determines the update parameter and the number of the re-identified edges. When the preset condition is met, the distance identifiers of all nodes to the sink point are updated to obtain a new distance identifier of all the nodes to the sink point, and the second residue is obtained according to the new distance identifier of all the nodes to the sink point. Quantity network.

For example, the algorithm selecting device determines to update the flow of the new active node, * when determining that the product of the number of re-identified edges and the update parameter is greater than the number of sides of the network flow graph and the product of the number, *, and number. The algorithm selection device may perform reverse width-first search by the sink point to update all the nodes to the sink point identifier, and after the distance identifier is updated, obtain the second residual network according to the updated distance identifier, and The number of key edges corresponding to the at least two algorithms is counted in the second residual network.

Optionally, the algorithm selecting device determines, when the data structure of the management active node does not have an active node, determining that the traffic of the sink is the maximum flow of the network, such that the at least two algorithms are calculated by executing the parallel algorithm in parallel. The network flow graph speeds up the acquisition of the maximum flow of the network, thereby improving the calculation efficiency.

Further, the algorithm selecting device acquires the total number of threads and performs the number of threads corresponding to at least two algorithms before the pre-streaming advancement by the at least two algorithms in the first residual network, and the algorithm selecting device is in the The pre-streaming advancement is performed in parallel by the at least two algorithms according to the corresponding number of threads in the first residual network.

The number of threads includes a ratio of the total number of threads to the number of algorithms of the at least two algorithms, that is, the number of threads configured by the algorithm selection device for each algorithm is the same.

For example, the code that the algorithm executes in parallel can be:

Pa ra l l e l - f or (the numbe r of a l gor i thms) {

Pa ra l l e l - f or (thread- grou [ i ] ) {

a 1 gor i thm_ i (); Paral lel-f or (the number of a 1 gor i thms) executes all algorithms in parallel, where the number of algorithms represents the number of algorithms, paral lel_f or (thread_group [i] ) executes algorithm i, where thread- Group [i] represents the thread group corresponding to algorithm i; algorithm_i () represents algorithm i.

It should be noted that the more the number of threads, the faster the calculation time of the corresponding algorithm and the higher the computational efficiency.

S104. The algorithm that determines, by the algorithm selection device, the maximum value of the number of the critical edges is an optimal network maximum flow algorithm of the network flow graph.

Specifically, the more the number of key edges corresponding to the algorithm, the less redundant operation of the algorithm. Therefore, the algorithm is determined to be the optimal network maximum flow algorithm corresponding to the network flow graph.

Further, after the algorithm corresponding to determining the maximum value of the number of key edges is the optimal network maximum flow algorithm, the algorithm selects the number of threads for the at least two algorithms according to the number of the key edges.

Specifically, after determining the optimal network maximum flow algorithm, more threads are allocated for the optimal network maximum flow, so that the algorithm reduces computation time and improves computation efficiency in the operation process, thereby improving the algorithm. Algorithm performance.

上述 Using the above-mentioned execution subject as an algorithm to select a device, the algorithm selection device pre-flows the network flow graph by operating multiple algorithms in the algorithm set, and after the pre-flow advancement, the number of key edges corresponding to each algorithm is the largest. The algorithm is determined as the optimal network maximum flow algorithm, so that in the process of obtaining the maximum flow of the network, the adaptive different network flow graphs can be dynamically adjusted to obtain an optimal performance algorithm, thereby improving the efficiency of the network maximum flow calculation.

An embodiment of the present invention provides a method for selecting an optimal network maximum flow algorithm, as shown in FIG. 2, including:

The S20 algorithm selects a device to obtain a network flow graph, and obtains a first residual network according to the network flow graph.

The residual network is obtained by transforming the network flow graph.

S202. The algorithm selection device acquires an active node according to the first residual network.

S203. The algorithm selection device acquires an algorithm and a data structure corresponding to the active node according to the algorithm template, and adds a fine-grained lock to the data structure of the network flow graph. The active node is a node whose inbound traffic is greater than the outgoing traffic. The algorithm template includes a data structure of the user-configured active active node and an algorithm configured by the user corresponding to the active node and a data structure of the network flow graph.

Specifically, the algorithm selection device puts the acquired active node into the data structure of the management active node.

For example, the algorithm template includes a main control part, providing a framework of the algorithm; the algorithm part, the user may configure at least two algorithms; the node management part, including the acquired active node and the user-configured corresponding management active node data Structure (such as bucket, FIFO-queue, and stack); update event, used to update the distance identifier of each node in the first residual network to the sink; update parameters, configured by the user; network flow graph Data structure for storing residual networks.

The algorithm part, the node management part and the update parameter can be configured by the user. For example, the above algorithm framework can be:

Whi le (act i veV) {

i = workse t. remove ();

Ver tex. compute (i);

Where activeV indicates the number of active nodes in the residual network, while (activeV) indicates that the number of active nodes in the residual network is not zero, the algorithm is executed, i = workset. remove () indicates that the active node Extracted from the node management part, the workset represents the data structure corresponding to the management active node configured by the user; vertex, compute (i) indicates that the surplus stream of the active node is operated by the algorithm corresponding to the active node.

Workse t:: bucket (i->d);

GLOB_UPDT-FREQ = 0.5; Where workset:: bucket (i->d) represents the data structure of the user-managed active node; vertex, computel (node* i) { ... workset. push ( j); } indicates the active node and the corresponding algorithm , GLOB_UPDT_FREQ = 0.5 indicates the user-configured update parameters. Parallel operations of multiple algorithms can be implemented by the above set of operational algorithms.

S204. The algorithm selection device obtains the operation algorithm set according to the algorithm and the data structure. The set of operational algorithms includes at least two algorithms.

S 205. The algorithm selection device acquires a corresponding active node from the data structure of the management active node according to the at least two algorithms.

5206. The algorithm selects a device to obtain a total number of threads, and configures a number of threads corresponding to at least two algorithms according to the total number of threads.

The number of threads includes a ratio of the total number of threads to the number of algorithms of the at least two algorithms.

S207. The algorithm selection device performs pre-flow advancement in parallel according to the corresponding number of threads in the first residual network by using the at least two algorithms.

For example, the code that the algorithm executes in parallel can be:

Paral lel-f or (the number of algorithms) {

Parallel-for (thread- grou [i] ) {

a 1 gor i thm_ i ();

Paral lel-f or (the number of a 1 gor i thms) executes all algorithms in parallel, where the number of algorithms represents the number of algorithms, paral lel_f or (thread_group [i] ) executes algorithm i, where thread- Group [i] represents the thread group corresponding to algorithm i; algorithm_i () represents algorithm i.

S208. After the pre-flow advancement, if there is a surplus flow in the adjacent node of the active node, the algorithm selecting device determines that the adjacent node is a new active node, and puts the new active node into the Manage active nodes, in the data structure.

S 209. The algorithm selection device acquires the number of edges that are re-identified when the pre-flow advancement is performed in parallel by the at least two algorithms, and acquires the update parameter according to the algorithm template. It should be noted that, if the algorithm selection device determines that there is no active node in the data structure of the active node after performing the pre-flow advancement, it determines that the traffic of the sink point is the maximum flow of the network, so that the at least The two algorithms calculate the network flow graph, which speeds up the acquisition of the maximum flow of the network, thereby improving the calculation efficiency.

S 21 0. The algorithm selecting device updates the distance identifiers of all the nodes to the sink point when determining that the update parameter and the number of the re-identified edges meet the preset condition, to obtain a new giant of all the nodes to the sink point. Off the logo.

For example, when determining that the product of the number of re-identified edges and the update parameter is greater than the product of the number of edges and the number of nodes of the network flow graph, the algorithm selecting device determines to update the distance identifier of all the nodes to the sink point, and the algorithm selecting device Specifically, the sinker performs reverse width priority search to update the distance identifier of all nodes to the sink point, and after the distance identifier is updated, obtains the second residual network according to the updated distance identifier, and in the second residual The number of key edges corresponding to the at least two algorithms is counted in the quantity network.

S 21 1. The algorithm selection device obtains the second residual network according to the new distance identifier. S 21 2. The algorithm selection device determines the number of key edges corresponding to the at least two algorithms in the second residual network.

S 21 3. The algorithm that determines the maximum value of the number of key edges by the algorithm selection device is an optimal network maximum flow algorithm of the network flow graph.

S214. The algorithm selection device reconfigures the number of threads for the optimal network maximum flow algorithm. Specifically, after determining the optimal network maximum flow algorithm, more threads are allocated for the optimal network maximum flow, so that the algorithm reduces computation time and improves computation efficiency in the operation process, thereby improving the algorithm. Algorithm performance.

The above scheme is adopted, so that in the process of obtaining the maximum flow of the network, the adaptive network flow graph can be dynamically adjusted to obtain an optimal performance algorithm, thereby improving the efficiency of the network maximum stream calculation.

It should be noted that, for the above method embodiments, for the sake of simple description, they are all expressed as a series of action combinations, but those skilled in the art should understand that the present invention is not limited by the described action sequence. Those skilled in the art should also appreciate that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention. An embodiment of the present invention provides an algorithm selection device 30, as shown in FIG. 3, including: The obtaining unit 31 is configured to obtain a network flow graph, obtain a first residual network according to the network flow graph, and obtain a set of operation algorithms.

The residual network is obtained by transforming a network flow graph, and the operational algorithm set includes at least two algorithms.

The processing unit 32 is configured to perform pre-flow advancement in parallel by the at least two algorithms in the first residual network to obtain a second residual network, and determine a key edge corresponding to the at least two algorithms after the pre-flow advancement The algorithm corresponding to the quantity, and determining the maximum value of the number of key edges in the second residual network is the optimal network maximum flow algorithm of the network flow graph.

Further, the acquiring unit 31 is further configured to obtain an active node according to the first residual network after obtaining the first residual network according to the network flow graph.

Specifically, the obtaining unit 31 puts the acquired active node into the data structure of the management active node.

The acquiring unit 31 is configured to obtain an algorithm and a data structure corresponding to the active node according to the algorithm template, and obtain the operation algorithm set according to the algorithm and the data structure, where the algorithm template includes a user-configured active node. The data structure and the user-configured algorithm corresponding to the active node and the data structure of the network flow graph.

For example, the algorithm template includes a main control part, providing a framework of the algorithm; the algorithm part, the user may configure at least two algorithms; the node management part, including the acquired active node and the user-configured corresponding management active node data Structure (such as bucket, FIFO-queue, and stack); update event, used to update the distance identifier of each node in the first residual network; update parameters, configured by the user, network flow graph data structure, Used to store residual networks.

Whi le (act iveV) {

i = workse t. remove ();

Ver tex. compute (i); where act iveV indicates the number of active nodes in the residual network, while (activeV) indicates that the number of active nodes in the residual network is not zero, the algorithm is executed, i = workset. Remove 0 means that the active node is taken out from the node management part, and the workset represents the data structure of the corresponding management active node configured by the user; vertex, compute (i) The surplus stream of the active node is operated by an algorithm corresponding to the active node.

Workse t:: bucket (i->d);

GLOB_UPDT-FREQ = 0.5;

Where workset: : bucket (i->d) represents the data structure of the user-managed active node; vertex, computel (node* i) { ... workset. push ( j); } indicates the active node and the corresponding algorithm , GL0B_UPDT_FREQ = 0.5 indicates the user-configured update parameters. Parallel operations of multiple algorithms can be implemented by the above set of operational algorithms.

Further, the processing unit 32 is further configured to: fine-grain the data structure of the network flow graph to prevent data competition, for example, prevent other algorithms other than the active node corresponding algorithm from reading and writing the active node. operating.

Optionally, the processing unit 32 is configured to: obtain, according to the at least two algorithms, a corresponding active node from the data structure of the active node, and perform the corresponding active node in parallel according to the at least two algorithms. The surplus flow advances to the meeting point through adjacent nodes.

It should be noted that the parallelism mentioned here means that in the process of executing one of the algorithms, another algorithm can be executed at the same time, and it is not limited to start execution at the same time.

Further, the processing unit is further configured to: after the surplus stream of the corresponding active node is advanced to the sink point by using an adjacent node whose distance identifier is lower than itself, if the neighboring node has the surplus flow, Then, the adjacent node is determined to be a new active node, and the new active node is placed in the data structure of the management active node.

The obtaining unit 31 is further configured to: before the pre-streaming advancement by the at least two algorithms in the first residual network, obtain a total number of threads, and the processing unit configures at least two algorithms according to the total number of threads. The number of threads, and pre-flow advancement in parallel by the at least two algorithms according to the corresponding number of threads in the first residual network.

The number of threads includes a ratio of the total number of threads to the number of algorithms of the at least two algorithms, that is, the number of threads configured by the algorithm selection device for each algorithm is the same. For example, the code that the algorithm executes in parallel can be:

Paral lel-f or (the number of algorithms) {

Parallel-for (thread- grou [i] ) {

a 1 gor i thm_ i ();

Optionally, the processing unit 32 is further configured to: when the pre-stream advancement is performed in parallel by the at least two algorithms, obtain the number of re-identified edges, obtain an update parameter according to the algorithm template, and determine the update parameter and When the number of the re-identified edges meets the preset condition, the distance identifiers of all the nodes to the sink point are updated to obtain a new distance identifier of all the nodes to the sink point, and the second distance identifier is obtained according to the new distance identifier. Residual network.

For example, when determining that the product of the number of re-identified edges and the update parameter is greater than the product of the number of edges and the number of nodes of the network flow graph, the algorithm selecting device determines to update the distance identifier of all the nodes to the sink point, and the algorithm selecting device Specifically, the sinking point performs reverse width first search to update the distance identifiers of all the nodes to the sink point, and after the distance is updated, obtains the second residual network according to the updated distance identifier, and in the second residual The number of key edges corresponding to the at least two algorithms is counted in the quantity network.

Further, the processing unit 32 is further configured to: after determining that the algorithm corresponding to the maximum value of the number of the critical edges is the optimal network maximum flow algorithm, reconfiguring the number of threads for the at least two algorithms according to the number of the critical edges.

After determining the optimal network maximum flow algorithm, more threads are allocated for the optimal network maximum flow, so that the algorithm reduces the operation time and improves the operation efficiency in the process of operation, thereby improving the algorithm performance of the algorithm.

Optionally, the obtaining unit 31 is further configured to: determine a data structure of the active node in the management When there is no active node, the flow of the sink is determined to be the maximum flow of the network. Thus, since the network flow graph is calculated by executing the at least two algorithms in parallel, the speed of acquiring the maximum flow of the network is accelerated, thereby improving the speed. Computational efficiency.

选择 Selecting the device by using the above algorithm, the algorithm selecting device performs pre-flow advancement of the network flow graph by using multiple algorithms in the algorithm set, and after the pre-flow advancement, determining the algorithm with the largest number of key edges corresponding to each algorithm as The optimal network maximum flow algorithm enables dynamic adaptation of different network flow graphs to obtain optimal performance algorithms in the process of obtaining the maximum flow of the network, thereby improving the efficiency of the network maximum flow calculation. The present invention provides an algorithm selection device 40. As shown in FIG. 4, the algorithm selection device 40 includes:

a processor (Rocessor) 41, a communication interface (42), a memory 43 and a communication bus 44; wherein the processor 41, the communication interface 42, and the memory 43 are completed by the communication bus 44 Communication with each other.

The processor 41 may be a multi-core CPU or a specific integrated circuit.

An ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement embodiments of the present invention.

The memory 43 is for storing program code, and the program code includes computer operation instructions and a network flow diagram. Memory 43 may contain high speed RAM memory and may also include non-volatile memory, such as at least "disk storage."

The communication interface 42 is configured to implement connection communication between the devices.

The processor 41 executes program code for acquiring a network flow graph, and obtains a first residual network according to the network flow graph, and acquires an operation algorithm set, where the at least two algorithms are performed in parallel in the first residual network. Pre-flow advancement, obtaining a second residual network, and determining the number of key edges corresponding to the at least two algorithms in the second residual network, and determining that the maximum value of the number of key edges corresponds to the network flow The optimal network maximum flow algorithm for the graph.

The set of operational algorithms includes at least two algorithms.

Optionally, the processor 41 is further configured to: acquire an active node according to the first residual network, acquire an algorithm and a data structure corresponding to the active node according to the algorithm template, and obtain the operation according to the algorithm and the data structure. Algorithm set.

The algorithm template includes a data structure configured by the user to manage the active node, an algorithm configured by the user corresponding to the active node, and a data structure of the network flow graph. Optionally, the processor 41 is further configured to add a fine-grained lock to the data structure of the network flow graph. Optionally, the processor 41 is configured to: obtain, according to the at least two algorithms, a corresponding active node from a data structure of the active node, and perform the corresponding active node in parallel according to the at least two algorithms. The surplus flow advances to the meeting point through adjacent nodes.

Optionally, the processor 41 is specifically configured to: if the neighboring node has the surplus flow, determine that the adjacent node is a new active node, and put the new active node into a management active In the data structure of the node.

Optionally, the processor 41 is further configured to: when the pre-streaming is performed in parallel by the at least two algorithms, obtain the number of edges that are re-identified, and obtain an update parameter according to the algorithm template, where the update parameter is determined and When the number of the re-identified edges meets the preset condition, the distance identifiers of all the nodes to the sink point are updated to obtain a new distance identifier of all the nodes to the sink point, and the second distance identifier is obtained according to the new distance identifier. Residual network.

Optionally, the processor 41 is further configured to: when there is no active node in the data structure of the management active node, determine that the traffic of the sink is the maximum flow of the network.

Optionally, the processor 41 is further configured to: obtain a total number of threads, and configure, according to the total number of threads, a number of threads corresponding to at least two algorithms, where the at least two algorithms are used according to the first residual network The corresponding number of threads is pre-flowed in parallel.

The number of threads includes a ratio of the total number of threads to the number of algorithms of the at least two algorithms;

Optionally, the processor 41 is further configured to reconfigure the number of threads for the at least two algorithms according to the number of the critical edges.

The above is only the specific embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily think of changes or substitutions within the technical scope of the present invention. It should be covered by the scope of the present invention. Therefore, the scope of the invention should be determined by the scope of the claims.

Claims

Rights request

1. A method for selecting the optimal network maximum flow algorithm, which is characterized by: obtaining a network flow graph, and obtaining the first residual network according to the network flow graph;

Obtain a set of operating algorithms, wherein the set of operating algorithms includes at least two algorithms; perform pre-flow advancement in parallel through the at least two algorithms in the first residual network to obtain a second residual network, and Determine the number of key edges corresponding to the at least two algorithms in the second residual network;

The algorithm corresponding to the maximum value of the number of critical edges is determined as the optimal network maximum flow algorithm of the network flow graph.

2. The method according to claim 1, characterized in that, after obtaining the first residual network according to the network flow graph, it further includes:

Obtain active nodes according to the first residual network;

The acquisition operation algorithm set includes:

Obtain the algorithm and data structure corresponding to the active node according to the algorithm template; wherein, the algorithm template includes a user-configured data structure for managing the active node and a user-configured algorithm corresponding to the active node and the Data structure of network flow graph;

The set of operating algorithms is obtained based on the algorithms and data structures.

3. The method according to claim 2, characterized in that the method further includes: adding fine-grained locks to the data structure of the network flow graph.

4. The method according to claim 1 or 3, wherein the parallel pre-flow advancement through the at least two algorithms in the first residual network includes:

Obtain the corresponding active node from the data structure that manages active nodes according to the at least two algorithms;

According to the at least two algorithms, the surplus flow of the corresponding active node is advanced to the sink through the adjacent node with a lower distance identifier than itself, wherein the surplus flow includes the input of the active node. The positive difference between flow rate and outflow rate.

5. The method according to claim 4, characterized in that, after advancing the surplus flow of the corresponding active node to the sink through the adjacent node, further comprising: if the corresponding If the surplus flow exists in a neighboring node, the neighboring node is determined to be a new active node, and the new active node is put into the data structure for managing active nodes.

6. The method according to claim 5, characterized in that, before obtaining the second residual network, it further includes: When pre-flow advancement is performed in parallel through the at least two algorithms, the number of re-identified edges is obtained;

Obtain update parameters according to the algorithm template;

When it is determined that the update parameters and the number of re-identified edges meet the preset conditions, update the distance identifiers from all nodes to the sink point to obtain new distance identifiers from all nodes to the sink point;

Obtaining the second residual network includes:

The second residual network is obtained according to the new distance identifiers from all nodes to the sink.

7. The method according to claim 5 or 6, characterized in that, when it is determined that there is no active node in the data structure for managing active nodes, it is determined that the flow of the sink point is the maximum flow of the network.

8. The method according to any one of claims 1 to 7, characterized in that, before performing pre-flow advancement in parallel through the at least two algorithms in the first residual network, the method further includes: :

Get the total number of threads;

The number of threads corresponding to the at least two algorithms is configured according to the total number of threads, wherein the number of threads includes the ratio of the total number of threads to the number of algorithms of the at least two algorithms; The parallel pre-flow advancement through at least two algorithms in the residual network includes:

In the first residual network, the at least two algorithms are used to perform pre-stream advancement in parallel according to the corresponding number of threads.

9. The method according to claim 8, characterized in that, after determining that the algorithm corresponding to the maximum value of the number of key edges is the optimal network maximum flow algorithm, it further includes:

Reconfiguring the number of threads for the at least two algorithms based on the number of critical edges.

10. An algorithm selection device, characterized by including:

The acquisition unit is used to obtain the network flow graph, obtain the first residual network according to the network flow graph, and obtain the operation algorithm set, wherein the operation algorithm set includes at least two algorithms;

A processing unit configured to perform pre-flow advancement through the at least two algorithms in parallel in the first residual network to obtain a second residual network, and determine the key corresponding to the at least two algorithms after the pre-flow advancement. The number of edges, and the algorithm corresponding to the maximum value of the number of key edges in the second residual network is determined to be the optimal network maximum flow algorithm of the network flow graph.

1 1. The device according to claim 1 0, characterized in that the acquisition unit is further configured to: after obtaining the first residual network according to the network flow graph, obtain the active node according to the first residual network point;

The acquisition unit is specifically configured to obtain the algorithm and data structure corresponding to the active node according to the algorithm template, and obtain the operation algorithm set according to the algorithm and data structure, wherein the algorithm template includes user configuration management The data structure of the active node, the algorithm configured by the user corresponding to the active node, and the data structure of the network flow graph.

12. The device according to claim 11, characterized in that the processing unit is further configured to add fine-grained locks to the data structure of the network flow graph.

13. The device according to claim 11 or 12, characterized in that the processing unit is specifically configured to obtain the corresponding active node from the data structure of the managed active node according to the at least two algorithms. node, and advance the surplus flow of the corresponding active node to the sink through the adjacent node with a lower distance identifier than itself according to the at least two algorithms in parallel, wherein the surplus flow includes the active node The positive difference between the inflow and outflow of a node.

14. The device according to claim 13, wherein the processing unit is further configured to: after advancing the surplus flow of the corresponding active node to the sink through the adjacent node, If the surplus flow exists in the adjacent node, determine the adjacent node as a new active node, and put the new active node into the data structure that manages active nodes. .

15. The device according to claim 14, wherein the processing unit is further configured to obtain the number of re-identified edges when pre-flow advancement is performed in parallel through the at least two algorithms, according to the The algorithm template obtains the update parameters, and when it is determined that the update parameters and the number of re-identified edges meet the preset conditions, it updates the distance identifiers from all nodes to the sink point to obtain the distance identifiers from all nodes to the sink point. A new distance identifier is obtained, and the second residual network is obtained based on the new distance identifier from all nodes to the sink point.

16. The device according to claim 14 or 15, wherein the processing unit is further configured to determine the sink point when it is determined that there is no active node in the data structure for managing active nodes. The traffic is the maximum flow of the network.

17. The device according to any one of claims 10 to 16, characterized in that the acquisition unit is further configured to perform prediction in parallel through the at least two algorithms in the first residual network. Before the stream is pushed, get the total number of threads;

The processing unit is also configured to configure the number of threads corresponding to the at least two algorithms, and In the first residual network, the at least two algorithms are used to perform pre-stream advancement in parallel according to the corresponding number of threads, where the number of threads includes the total number of threads and the number of algorithms of the at least two algorithms. ratio.

18. The device according to claim 17, wherein the processing unit is further configured to, after determining that the algorithm corresponding to the maximum number of key edges is the optimal network maximum flow algorithm, based on the The number of critical edges is the number of reconfiguration threads of the at least two algorithms.