JP2016046785A - Cache server selection device, distributed cache system, and cache server selection method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a device and method for selecting a cache server for transmitting a requested content from a plurality of cache servers, by considering capacity and a congestion situation of the whole network, and a distributed cache system.SOLUTION: A cache server selection device receives a request from a client; specifies, from among a plurality of cache servers, at least one cache server storing data corresponding to the received request; specifies, on the basis of topology information, for each of the specified cache servers, at least one bundle of paths satisfying a predetermined condition from among at least one path capable of reaching from the device to the cache server; calculates network traffic capacity of each of the specified at least one bundle of paths for each of the specified cache servers; and selects, on the basis of the calculated capacity of the bundle of the respective paths, a cache server for transmitting data to the client, from among the specified cache servers.SELECTED DRAWING: Figure 6

Description

  The present invention relates to a cache server selection device that selects a cache server that transmits data to a client from a plurality of cache servers that hold data.

  With the global proliferation of broadband Internet in households and mobiles, the number of Internet services that require wide bandwidth is also increasing. This increases traffic.

  While innovation has increased the available bandwidth and reduced the delay caused by network infrastructure equipment, the demand for traffic is increasing rapidly. In particular, in a network carrier with many users, the traffic demand may exceed the capacity of the network. Based on the generally accepted observation that Internet traffic such as web and video follows Zip's law, the cache system is a mechanism that effectively reduces traffic.

  The cache system reduces the load on the origin server that provides data and the amount of use of the network between the cache server and the origin server, and at the same time, shortens the transmission time to the client. Services that require large amounts of data to be transmitted, such as high-resolution video streaming services, already have high storage capacity and bandwidth, so in large carrier-grade networks, the number and demand of requests Both storage capacity and bandwidth are expected to be very high. In order to increase the robustness and scalability of such a system, it is desirable to distribute the storage capacity and bandwidth using a plurality of cache servers. A distributed cache system is known as such a cache system.

  In a distributed cache system, the requested content is held by a plurality of cache servers, thereby shortening the average distance to the client that requested the content and increasing the throughput available for content transmission. In such a system, selecting one cache server that transmits content from a plurality of distributed cache servers directly uses the resources used for content transmission and the network used for content transmission. Closely related to various impacts on possible resources. Based on a comparison of the availability of network resources and computing resources for each cache server, or a comparison of the end-to-end distance (e.g. latency or hop count) of network resources between the client and each cache server, A cache server that transmits content is determined. Selection based on these evaluation measures can reduce the time taken to transmit the content to the client, and can achieve simple load balancing of the resources of the cache server.

  As a technique related to the distributed cache system, there is US Pat. No. 7,890,656 (Patent Document 1). The gazette of Patent Document 1 states that “a plurality of distribution servers in a transmission system including a plurality of distribution server side routers that communicate with a plurality of distribution servers that distribute content and a client side router that communicates with a client that receives content”. A database that holds the server load state of the server and individual link load states for a plurality of distribution paths between a plurality of distribution server side routers and client side routers, and the server load state and each individual link load held in the database And a path control unit that determines a minimum load state distribution path that minimizes the load state among a plurality of distribution paths based on the sum of the state and the state ”.

  Further, there is US Pat. No. 7,583,677 (Patent Document 2) as a technique related to multipath routing. According to the publication of Patent Document 2, “Generally, the present invention is directed to a technique for dynamically adjusting a network traffic load between a plurality of paths through a computer network. This technique is based on a dynamically measured path. Distributes and redistributes the flow of network packets between different paths based on the bandwidth and load of each flow, maintaining the QoS (Quality Of Service) bandwidth required for the flows when the flows are distributed. It is described.

  There is a technique called max flow processing for grasping the situation of the entire network regarding the throughput capacity and obtaining the maximum flow that can flow between certain points. In the MaxFlow process, by modeling the communication network as a flow network, the resources of the entire network available for the routing layer are calculated together with the capacity indicating the link bandwidth and the flow indicating the current usage. If the various available networks are fully used, the max flow between the source and the server sink shows the maximum available bandwidth.

  A technology related to max flow processing is “A New Approach to the Maximum-Flow Problem” (Non-Patent Document 1). Non-Patent Document 1 describes a method for efficiently calculating the maximum flow of a flow network in a distributed environment. When the maximum flow of each cache server is calculated, this evaluation measure can be used to select the cache server with the path having the maximum available capacity, and as described in Non-Patent Document 1. This technology can minimize network congestion.

US Pat. No. 7,890,656 US Pat. No. 7,583,677

Andrew Goldberg and Robert Tarjan, “A New Approach to the Maximum-Flow Problem”, Journal of the ACM, Vol. 35, 1988 Rowstron, A. & Druschel, P. “Pastry: Scalable, distributed object location and routing for large-scale peer-to-peer systems”, IFIP / ACM International Conference on Distributed Systems Platforms (Middleware), 2001, 329-350 Dabek, F .; Cox, R .; Kaashoek, F. & Morris, R. “Vivaldi: a decentralized network coordinate system”, SIGCOMM Comput. Commun. Rev., ACM, 2004, 34, 15-26

  The technique described in Patent Document 1 calculates the shortest path (end-to-end) from a client-side router to a plurality of content server-side routers, and determines a route used for content transmission based on the calculation result. The content server side router as an end point is limited to the router with which the client side router communicated in the past. For this reason, the technique described in Patent Document 1 cannot select a route used for content transmission in consideration of the entire network. Further, since the technique described in Patent Document 1 determines a route used for content transmission based on information collected in advance, it cannot cope with a network in which the path configuration rapidly changes.

  The multipath routing described in Patent Document 2 is a process executed in the third layer (networking layer). However, sub-optimal selection of transmissions to the endpoint can cause traffic path congestion because the choice of traffic at the routing layer is not diversified. This occurs because there is no grasp of the overall network usage during the selection process in the distributed system.

  In addition, the technique described in Non-Patent Document 1 can grasp the situation of the entire network. However, the messaging overhead during computation is complicated. As the number of nodes increases, traffic increases in a distributed cache system. This increases the time and messaging overhead required to compute the maxflow. The technique described in Non-Patent Document 1 cannot cope with a distributed cache system having a large number of nodes.

  An object of the present invention is to provide a cache server selection device that selects a cache server that transmits content in consideration of the situation of the entire network even in a distributed cache system having a large number of nodes.

  A representative example of the present invention is a cache server selection device connected to a client and a plurality of cache servers, the plurality of cache servers constructing a network and storing topology information of the network , Receiving a request from the client, identifying at least one cache server storing data corresponding to the received request from the plurality of cache servers, and for each of the identified at least one cache server, Based on the topology information, specify at least one path satisfying a predetermined condition from all paths reachable from the cache server selection device to the specified at least one cache server. For each cache server Calculating a capacity of each specified at least one path, and selecting a cache server that transmits the data to the client from the specified at least one cache server based on the calculated capacity of each path. Features.

  The effects obtained by the representative ones of the inventions disclosed in the present application will be briefly described as follows. That is, even a distributed cache system with a large number of nodes can provide a cache server selection device that selects a cache server that transmits content in consideration of the situation of the entire network.

  Problems, configurations, and effects other than those described above will be clarified by the following description of embodiments.

It is a figure of the structure of the communication system of Example 1. FIG. 1 is a schematic explanatory diagram of a communication system according to a first embodiment. FIG. 2 is a hardware configuration diagram of the cache server according to the first embodiment. 3 is a flowchart of processing executed by a cache server serving as a guardian according to the first exemplary embodiment. It is a flowchart of the distributed maxflow process performed by the guardian of Example 1. FIG. 4 is a flowchart of subgraph initialization processing according to the first embodiment. 6 is an explanatory diagram of a data structure of a subgraph according to Embodiment 1. FIG. It is a flowchart of the process performed by the cache server which is not a guardian when the push message of Example 1 is received.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  Example 1 will be described with reference to FIGS.

  In a distributed cache system having a plurality of cooperating cache servers, the network capacity between the cache server receiving the request and the plurality of cache servers holding the requested content is calculated, and the network capacity is maximum. The cache server is selected as the cache server that transfers the content.

  This reduces the amount of computation required and optimizes the network capacity used by the routing infrastructure of the underlay network without limiting system scalability.

  FIG. 1 is an explanatory diagram of a configuration of the communication system 100 according to the first embodiment.

  The communication system 100 includes a carrier network 101 and a content network 102. The content network 102 is a network different from the carrier network 101. The carrier network 101 is connected to the content network 102 via a network link 201.

  The network infrastructure of the carrier network 101 includes routers 340A and 340M, for example. In the carrier network 101, there are a client 301 and cache servers 360A to 360N. Within the content network 102 is a content server 321.

  The client 301 transmits a content request to the content server 321. The content request passes through the cache server 360A closest to the client 301. The cache server 360A is a part of the cooperative network of the cache servers 360A to 360N, and these cache servers 360A to 360N use the routers 340A and 340M when transferring data.

  When the cache server 360 </ b> A receives the content request transmitted by the client 301, the cache server 360 </ b> A specifies at least one cache server that holds content corresponding to the received content request. The cache server 360A specifies at least one path that satisfies a predetermined condition from among the paths that can reach each specified at least one cache server. Then, the cache server 360A calculates the capacity of each identified path, and selects a cache server that transfers the requested content to the client 301 based on the calculated capacity of each path. For example, the cache server 360A specifies a plurality of cache servers that hold the content corresponding to the received content request, specifies a reachable path for each of the specified cache servers, and specifies each of the specified plurality of paths. The path capacity is calculated, the path having the maximum calculated path capacity is selected, and the cache server to which the selected path reaches is selected as the cache server that transfers the content. Then, the cache server 360A transmits a content request to the selected cache server via the selected path. The cache server that has received the request transmitted by the cache server 360 </ b> A transmits the content corresponding to the received request to the client 301. This transmission may pass through the cache server 360A, and the cache server 360A may hold this content at that time.

  When the cache server 360A cannot identify at least one cache server that holds content corresponding to the received content request, that is, when there is no cache server that holds content corresponding to the received content request, the received content request Is transmitted to the content server 321. When the content server 321 receives this content request, the content server 321 transmits content corresponding to the received content request to the client 301 via the carrier network 101. This transmission may pass through the cache server 360A, and the cache server 360A may hold this content at that time.

  FIG. 2 is a schematic explanatory diagram of the communication system 100 according to the first embodiment.

  A communication system 100 illustrated in FIG. 1 includes a distributed cache system 500. The distributed cache system 500 includes, for example, a plurality of cache servers 360A to 360D, a client 301, and a content server 321. The client 301 transmits a content request 401 that requests the content server 321 to transmit a certain content. The cache server 360A closest to the client 301 receives the content request 401. This cache server 360A is called a guardian.

  The client 301 may use the guardian as a proxy that transfers the content request 401, or the network is set so that the guardian is used like a transparent proxy that transfers the content request 401 without setting on the client 301 side. Also good.

  At least one of the cache servers 360 </ b> A to 360 </ b> D of the distributed cache system 500 can hold the content requested by the content request 401. The guardian specifies all the cache servers that hold the content corresponding to the content request 401 using the query function of the overlay network in which the cache servers 360A to 360D participate. All cache servers that hold content corresponding to the content request 401 are called delegates.

  At least one of the cache servers 360 </ b> A to 360 </ b> D holds a delegate list in which a list of content identifiers and a list of delegate identifiers that hold content identified by the content identifiers are registered. A cache server that holds a delegate list for a content request identified by an identifier of a certain content is called a home.

  When the guardian receives the content request 401, the guardian transmits a delegate query message 411 including an identifier of the content included in the received content request 401 to the adjacent cache server in order to search for a home cache server. When the adjacent cache server receives the delegate query message 411, it determines whether or not it holds a delegate list for the identifier of the content included in the received delegate query message 411.

  When it is determined that the adjacent cache server holds the delegate list, the adjacent cache server transmits a delegate query response 412 including the delegate list to the guardian as a response to the received delegate query message 411. On the other hand, when it is determined that the adjacent cache server does not hold the delegate list, the received delegate query message 411 is transmitted to another adjacent cache server. Details of the search method of the cache server holding the delegate list are described in “Pastry: Scalable, distributed object location and routing for large-scale peer-to-peer systems” (Non-patent Document 2). The contents described in Non-Patent Document 2 are incorporated in this application.

  In FIG. 2, the home is the cache server 360C. When the cache server 360C receives the delegate query message 411, the cache server 360C transmits a delegate query response 412 including a delegate list to the cache server 360A that is the guardian. When a plurality of delegate identifiers corresponding to the content identifiers included in the delegate query message 411 are registered in the delegate list included in the delegate query response 412, the guardian displays the virtual coordinate system 803 shown in FIG. 4. Is used to exclude all delegates whose network distance is greater than or equal to a predetermined value. The virtual coordinate system 803 maintains an approximate value of the network distance between cache servers. When a certain cache server communicates with another cache server, the virtual coordinate system 803 is updated. For example, the network distance is based on latency between cache servers or the number of hops. Details of the virtual coordinate system are described in “Vivaldi: a decentralized network coordinate system” (Non-patent Document 3). The contents described in Non-Patent Document 3 are incorporated in this application.

  The guardian uses the topology information 805 shown in FIG. 4 to calculate a subgraph 806 for each delegate shown in FIG. The topology information 805 of each cache server is set a priori by a network administrator or acquired using a topology discovery function and a network management function. The topology information 805 includes link capacity and link usage between the cache server and the router or between different routers. In different embodiments, other information (eg, latency or packet loss rate) may be included in the topology information 805.

  Each subgraph 806 includes only nodes on the path between the guardians that satisfy a predetermined condition and construct a certain form of corridor and each delegate. The corridor is identified based on the network distance between the guardian and each delegate.

  Then, the guardian executes the distributed maxflow process for each delegate using the subgraph 806 in order to reduce the calculation amount of the distributed maxflow process. Since the non-patent document 1 describes the distributed max flow, the description is omitted. In the present embodiment, the distributed maxflow process is executed for a path satisfying a predetermined condition among all paths between the guardian and each delegate. As a result, the amount of computation of the distributed maxflow process and the number of messages communicated during the computation of the distributed maxflow process can be reduced, and the time taken for the distributed maxflow process can be shortened. Therefore, the distributed maxflow process can be applied to a distributed cache system having a large number of nodes, and the guardian can select a cache server that transmits content in consideration of the situation of the entire network.

  For example, in FIG. 2, the delegates are the cache servers 360B and 360D. Distributed maxflow messages 421 and 422 are communicated between the guardian and the cache servers 360B and 360D in order to execute the distributed maxflow processing. Based on the result of the distributed maxflow process, the guardian selects the delegate that the corridor having the maximum capacity reaches as the cache server that transmits the content to the client 301. In FIG. 2, the cache server 360 </ b> D is selected as the cache server that transmits the content to the client 301.

  Then, the guardian transmits the content request 401 transmitted by the client 301 to the cache server 360D as a message 431. When the cache server 360D receives the message 431, the cache server 360D transmits a response 432 including content corresponding to the content identifier included in the message 431 to the guardian. When receiving the response 432, the guardian transmits the received response 432 to the client 301 as the response 402. In this case, the guardian may store the content included in the received response 432.

  If there is no cache server that holds content corresponding to the content identifier corresponding to the content request 401, the guardian directly transmits the content request 401 as the content request 441 to the content server 321. When the guardian receives the content server response 442 transmitted by the content server 321, the guardian transmits the received content server response 442 to the client 301 using the response 402. In this case, the guardian may store the content included in the received content server response 442.

  In this embodiment, as described in Non-Patent Document 2, the cache server is a part of a virtual peer-to-peer overlay network that provides an address function, a discovery function, and a routing function. The cache servers can communicate with each other using a peer-to-peer overlay network and can communicate directly with each other (eg, without a peer-to-peer overlay network). Each cache server grasps the topology of the underlay network. The topology of the underlay network may be introduced during the initial setup of each cache server by the network administrator. Also, the topology of the underlay network includes the maximum capacity of the network link between the nodes. All cache servers use the virtual coordinate system described in Non-Patent Document 3. Each cache server can acquire resource information of other adjacent nodes and network components. In particular, the resource information includes a usage amount of a link connected to the node.

  In a peer-to-peer system, each node of a distributed cache system is equal in terms of workflow, responsibility, and communication, and thus each node is referred to as a “peer”. In widely deployed peer-to-peer systems, the use of a distributed hash table (DHT) allows the discovery of other nodes by using hash values as identifiers and using routing mechanisms within peer-to-peer networks. . Non-Patent Document 2 describes an overlay network on a routing infrastructure constructed by a peer. This overlay network can provide efficient content discovery and functional routing using mathematical hash values as identifiers for peers and content stored at peers.

  In the virtual coordinate system described in Non-Patent Document 3, each node places itself and other nodes in a virtual Euclidean space coordinate system, and the network latency measured when directly communicating and other nodes received. The coordinates of itself and other nodes are adjusted based on the information regarding the coordinates of the node.

  FIG. 3 is a hardware configuration diagram of the cache server according to the first embodiment.

  The cache server according to this embodiment includes a central processing unit (CPU) 1501, a primary memory 1502, a network interface controller (NIC) 1504, and a secondary memory 1505. These are connected to each other via a communication bus 1503. The minimum components required for a cache server are illustrated in FIG. For example, other configurations such as other network interface controllers may be connected to the communication bus 1503.

  During the operation of the cache server, the operation instruction of the cache server and the configuration parameters required at the time of execution are stored in the primary memory 1502. When the CPU 1501 executes the program stored in the primary memory 1502, the operation of the cache server is realized, and the CPU 1501 can access all the components connected to the communication bus 1503.

  Secondary memory 1505 provides storage space for data. The secondary memory 1505 is a non-volatile memory, and the primary memory 1502 is a volatile memory. The network interface controller 1504 implements communication with external components of the cache server itself.

  The hardware configuration of the cache servers 360A to 360N shown in FIG. 1 is the same as the hardware configuration of the cache server shown in FIG.

  FIG. 4 is a flowchart of processing executed by the cache server serving as the guardian of the first embodiment.

  The cache server closest to the client executes the task as a guardian. The guardian receives a content request from the client.

  When receiving the content request (1001), the cache server serving as the guardian extracts the requested content position 801 from the received content request. The guardian then requests a delegate list for the content identifier corresponding to the requested content from the virtual peer-to-peer overlay network (1002). The guardian receives a delegate list 802 from a virtual peer-to-peer overlay network. The delegate list 802 includes an identifier of a cache server that holds the requested content. If there is no cache server holding the requested content, the delegate list 802 does not include any identifiers of other cache servers. In other words, the delegate list 802 may be empty. The guardian may add the identifier of the content server 321 to the delegate list 802.

  If a delegate with a long distance to the guardian is chosen to send the requested content to the client, the content being sent will go through many network links and many nodes and network links will be affected, possibly Cause congestion. For this reason, all cache servers whose network distance from the guardian is equal to or greater than a predetermined value are deleted from the delegate list 802. In order to prevent distributed maxflow processing from being performed for those cache servers, the guardian satisfies the predetermined distance function using the current state of the virtual coordinate system 803 and the delegate list 802 (original). The delegate identifiers of all the cache servers included in the delegate list 802 are deleted (1003), and a reduced delegate list 804 is generated.

  In the processing of step 1003, the cache server sorts the delegate identifiers included in the delegate list 802 in the order of the network distance of each delegate of the virtual coordinate system 803, and identifies the identifiers of all delegates whose network distance is greater than the selected delegate. It may be deleted from the delegate list 802.

  In one example, a delegate with a network distance of 10 may be selected. In another example, a delegate corresponding to a value obtained by dividing the number of all delegates included in the delegate list 802 by 3 in the position of the delegate list 802 may be selected.

  The cache server selects one delegate from the reduced delegate list 804 (1004), and executes distributed maxflow processing using the topology information 805 for the selected delegate (1005). If there is a delegate in the reduced delegate list 804 for which the distributed maxflow processing 1005 has not been executed, the distributed maxflow processing is executed for this delegate (1005). If the distributed maxflow process is executed for all the delegates included in the reduced delegate list 804, the cache server executes the process of step 1006. In the distributed maxflow process, the cache server calculates the capacity of the path from the guardian to the delegate selected in step 1004. Details of the distributed maxflow processing will be described with reference to FIG.

  The distributed maxflow process for each delegate returns a maxflow value. The maximum flow value is a value indicating the capacity of all paths connecting from the guardian to the delegate selected in the processing of step 1004. The cache server registers the max flow value in the max flow list 807. The cache server performs distributed maxflow processing on all delegates included in the reduced delegate list 804, and then selects a delegate from the maxflow list 807 (1006). In one example, the cache server selects the delegate with the maximum max flow value as the optimal delegate. Then, the cache server transfers the content request received in the process of step 1002 to the delegate selected in the process of step 1006 (1007), and ends the process. The content request transferred in step 1007 corresponds to the message 431 shown in FIG.

  When the delegate receives the content request transferred in step 1007, the delegate transmits a response including the requested content to the guardian. The guardian forwards the received response to the client. The guardian may store the content included in the received response in the secondary memory 1505. In this case, the guardian updates the delegate list of the virtual peer-to-peer overlay network to add the relationship between the identifier of the content stored by the guardian and the identifier of the cache server serving as the guardian.

  In one example, if the requested content has been deleted from the delegate secondary memory 1505, the response sent by the delegate may include an error code. In this case, the guardian may select another delegate in the process of step 1006 after deleting the delegate that transmitted the response including the error code from the reduced delegate list 804.

  If all the delegates in the reduced delegate list 804 send a response including an error code, the guardian transfers the content request to the content server 321. The content server 321 transmits a response including the requested content to the client 301 via the guardian. When the guardian receives the response transmitted by the content server 321, the guardian may store the content included in the received response in the secondary memory 1505. The guardian also needs to update the virtual peer-to-peer overlay network delegate list.

  Next, the distributed maxflow process will be described with reference to FIGS. FIG. 5 is a flowchart of the distributed maxflow process executed by the guardian according to the first embodiment.

  First, the cache server serving as the guardian generates a UUID (Universal Unique Identifier) for transmission of the subgraph 806 (1101). For example, the cache server may generate a UUID using a hash function in order to prevent duplication between UUIDs. In one example, the input of the hash function is a guardian IP address and an identifier of the guardian CPU 1501.

  Then, the cache server executes a subgraph initialization process for generating the subgraph 806 (1102). Details of the subgraph initialization processing will be described with reference to FIG. Details of an example of the data structure of the subgraph 806 will be described with reference to FIG.

  Then, the cache server executes a normal initialization process (1103) of the distributed maxflow process, and ends the distributed maxflow process. In the process of step 1103, the cache server calculates the capacity of the path included in the limited subgraph 806 from the guardian to the delegate selected in the process of step 1004. The path for which the capacity is calculated is included in the limited subgraph 806. In the processing of step 1103, the cache server transmits the push message including the limited subgraph 806 to the cache server included in the limited subgraph 806. Details of the path capacity calculation process are described in Non-Patent Document 1.

  FIG. 6 is a flowchart of the subgraph initialization process according to the first embodiment.

  First, the guardian converts the topology information 805 into an acyclic graph 808. The guardian then calculates the shortest network distance 809 between the guardian and the selected delegate based on the graph 808 (1201). In one example, the cache server calculates the shortest path using the Dijkstra's shortest path priority algorithm.

  Then, the cache server calculates the limited length using the shortest path network distance 809 and the limiting element 810 calculated in the processing of Step 1201 (1202). For example, the limiting element 810 is stored in the primary memory 1502 of the cache server. In one example, the cache server calculates the limit length by adding the limit element 810 to the network distance 809. In another example, the cache server may calculate the limit length by adding a limit element 810 to the network distance 809.

  Then, the cache server has a path length that is not less than the shortest path network distance 809 among all the paths from the guardian to the delegate in the graph 808 and not more than the limited length calculated in the processing of step 1202. A certain path is specified (1203). The cache server registers the path specified in the process of step 1203 in the path list 811. The path list 811 is used to generate a subgraph.

  Then, the cache server generates a subgraph 806 using the path list 811 (1204), and ends the subgraph initialization processing. The subgraph 806 includes nodes and edges of the graph 808 and the path list 811.

  FIG. 7 is an explanatory diagram of a data structure of the sub-graph 806 according to the first embodiment. The subgraph 806 is stored in the primary memory 1502 of the cache server, and is transmitted to other cache servers by a push message.

  The data structure of the subgraph 806 includes a UUID field 1401, an edge number field 1402, and edges 1403A to 1403N. Each edge field 1403 </ b> A to 1403 </ b> N includes a start point node field 1404 and an end point node field 1405.

  In the UUID field 1401, the UUID generated in the process of step 1101 is registered. In the example where the hash value is used for the UUID, the capacity of the UUID is fixed. As a result, the value of the UUID field 1401 is easily decoded. In the edge number field 1402, a value indicating the number of edges included in the edge set E ″ of the subgraph 806 is registered. In an example, the capacity of the edge number field 1402 may be fixed. The number of edges written in the number field 1402 is easily decoded, and the number of edges 1403A to 1403N included in the subgraph 806 is the same as the number indicated by the value registered in the edge number field 1402. The start node field 1404 contains The identifier of the cache server, router, or other network device that is the start point of the edge is registered, and the identifier of the cache server, router, or other network device that is the end point of the edge is registered in the end node field 1405. In one example, a start node 1404 and an end node Identifier registered in 405 is the IP address of the cache servers and routers and other network devices.

  Next, processing executed by a non-guardian cache server when a push message including the subgraph 806 is received will be described with reference to FIG. FIG. 8 is a flowchart of processing executed by a non-guardian cache server when the push message of the first embodiment is received.

  First, the cache server receives a push message including a new UUID transmitted by the guardian (1301). The new UUID is different from the UUID included in the push message received so far by the cache server. If the UUID included in the received push message is not a new UUID, the push message is ignored. In an example, the cache server stores a list in which the UUID included in the received push message is registered in the secondary memory 1505. When the UUID included in the received push message is not in the list, the cache server adds the UUID included in the received push message to the list.

  Then, the cache server parses the data structure of the subgraph 806 included in the received push message (1302). Then, the cache server updates or generates the graph 808 using the subgraph 806 (1303).

  In one example, the cache server uses the subgraph 806 to generate a temporary graph 808 that is stored in the primary memory 1502. Further, the cache server generates a new graph 808 by using the combination of the topology information 805 that may be slightly different in each cache server and the subgraph 806 included in the received push message, and updates the topology information 805. To do. As a result, the cache server can add a node that is not included in the topology information 805 or can delete a node that is not included in the updated graph 808.

  Then, the cache server executes a normal initialization process of the distributed maxflow process (1304), and ends the process when the push message is received. Details of the processing in step 1304 are described in Non-Patent Document 1.

  In this embodiment, the guardian specifies all paths that satisfy a predetermined condition from at least one path between the guardian and the delegate. The guardian then performs distributed maxflow processing on these paths and calculates the maxflow capacity up to each delegate. This reduces the number of nodes used by the distributed maxflow process, thereby reducing the number of messages communicated during the maxflow computation. Therefore, this embodiment can provide a cache server selection device that selects a cache server that transmits content in consideration of the situation of the entire network even in a distributed cache system having a large number of nodes.

  When content is transmitted by a delegate with a long network distance from the guardian, network congestion is caused. For this reason, in this embodiment, the guardian focuses only on a delegate that is equal to or smaller than a predetermined network distance to the guardian, and further focuses only on a path that satisfies a predetermined condition based on the distributed maxflow calculation. This effectively eliminates the number of messages communicated to calculate MaxFlow capacity, since unnecessary long paths that cause congestion in unnecessary networks are not considered during the calculation.

  In one example, the condition for selecting a path between the guardian and any one delegate may be related to the shortest path distance between the guardian and the delegate. In other examples, other possible parameters may be included. This reduces, for example, the number of paths considered in relation to conditions related to the length of the shortest path, so that messages communicated during the computation of the capacity of the max flow between the guardian and the delegate The number can be reduced efficiently.

  In this embodiment, the guard server serving as the guardian transmits the subgraph information of all paths to each cache server satisfying a predetermined condition to a plurality of cache servers. The plurality of cache servers update their topology information based on the received subgraph information that may be up-to-date and may include information that each cache server does not know. As a result, even when each cache server is distributed and manages its own topology information, the topology information as a whole is collected into an accurate topology.

  In the first embodiment, the cache server serving as the guardian transmits a push message including the subgraph 806 in the process of step 1204 illustrated in FIG. In the second embodiment, a cache server serving as a guardian transmits a push message including TTL (Time-to-Live). When receiving the push message, the cache server that is not the guardian does not execute the processes of steps 1302 and 1303 shown in FIG. 8, but executes the process of step 1304.

  According to the present embodiment, the guardian uses the limited length calculated in the process of step 1202 shown in FIG. 6 as the TTL value. When transmitting the push message, the cache server writes the TTL value in the edge number field 1402 shown in FIG. The push message transmitted by the cache server serving as the guardian does not include the edge fields 1403A to 1403N.

  When the cache server that is not the guardian receives the push message, the cache server subtracts a predetermined value from the TTL included in the received push message. For example, this value is 1. A non-guardian cache server ignores a received push message if the recalculated TTL is zero. If the subtracted TTL is not 0, the non-guardian cache server executes the distributed maxflow process described in Non-Patent Document 1.

  In this example, nodes far from the guardian and having a TTL of 0 are not considered in the calculation and are not used during transmission after the delegate is selected. Therefore, even if the transmitted push message does not include the limited subgraph 806, Network congestion can be prevented.

  In addition, this invention is not limited to an above-described Example, Various modifications are included. For example, the above-described embodiments have been described in detail for easy understanding of the present invention, and are not necessarily limited to those having all the configurations described. Further, a part of the configuration of a certain embodiment can be replaced with the configuration of another embodiment, and the configuration of another embodiment can be added to the configuration of a certain embodiment. Further, it is possible to add, delete, and replace other configurations for a part of the configuration of each embodiment.

  Each of the above-described configurations, functions, processing units, processing means, and the like may be realized by hardware by designing a part or all of them with, for example, an integrated circuit. Each of the above-described configurations, functions, and the like may be realized by software by interpreting and executing a program that realizes each function by the processor. Information such as programs, tables, and files that realize each function can be stored in a recording device such as a memory, a hard disk, or an SSD (Solid State Drive), or a recording medium such as an IC card, an SD card, or a DVD.

  In addition, control lines and information lines are those that are considered necessary for explanation, and not all control lines and information lines on the product are shown. In fact, almost all configurations are connected to each other. You may think that it is.

DESCRIPTION OF SYMBOLS 100 Communication system 101 Carrier network 102 Content network 201 Network link 301 Client 321 Content server 340A, 340M Router 360A-360N Cache server

Claims (15)

A cache server selection device connected to a client and a plurality of cache servers,
The plurality of cache servers construct a network,
Storing topology information of the network;
Receiving a request from the client;
Identifying at least one cache server that stores data corresponding to the received request from the plurality of cache servers;
For each of the specified at least one cache server, at least one path satisfying a predetermined condition from all paths reachable from the cache server selection device to the specified at least one cache server is determined as the topology. Identify based on information,
Calculating the capacity of each identified at least one path for each identified at least one cache server;
A cache server selection apparatus, wherein a cache server that transmits the data to a client is selected from the specified at least one cache server based on the calculated path capacity. The cache server selection device according to claim 1,
At least one path satisfying the predetermined condition is a path having a predetermined relationship with the cache server selection apparatus. The cache server selection device according to claim 2,
The path having a predetermined relationship with the cache server selection apparatus is a path that reaches a predetermined distance from the cache server selection apparatus to the specified at least one cache server. The cache server selection device according to claim 3,
The shortest path that reaches from the cache server selection device to the specified at least one cache server with the shortest distance among all the paths that can be reached from the cache server selection device to the specified at least one cache server. Identify
A path that reaches the specified at least one cache server from the cache server selecting device at a predetermined distance is determined from a distance of the specified shortest path from the cache server selecting device to the specified at least one cache server. A cache server selection device characterized in that the path reaches a range. The cache server selection device according to claim 1,
The plurality of cache servers store topology information of the network,
Based on the topology information, identify a plurality of nodes on a path satisfying the specified predetermined condition, and links constituting the path,
A cache server selection device that transmits subgraph information including information regarding the specified nodes and the specified link to the plurality of cache servers in order to cause the plurality of cache servers to update the topology information. . A distributed cache system comprising a client and a plurality of cache servers,
The plurality of cache servers construct a network,
The plurality of cache servers store topology information of the network,
The guardian that is the cache server that received the request from the client,
Identifying at least one cache server that stores data corresponding to the received request from the plurality of cache servers;
For each of the specified at least one cache server, at least one path satisfying a predetermined condition among all paths reachable from the guardian to the specified at least one cache server is based on the topology information. Identify
Calculating the capacity of each identified at least one path for each identified at least one cache server;
A distributed cache system, wherein a cache server that transmits the data to a client is selected from the specified at least one cache server based on the calculated capacity of each path. The distributed cache system according to claim 6,
The distributed cache system, wherein at least one path satisfying the predetermined condition is a path having a predetermined relationship with the guardian. The distributed cache system according to claim 7,
The distributed cache system according to claim 1, wherein the path having a predetermined relationship with the guardian is a path that reaches a predetermined distance from the guardian to the specified at least one cache server. The distributed cache system according to claim 8,
The guardian selects a shortest path that reaches the shortest distance from the guardian to the at least one specified cache server among at least all paths that can be reached from the guardian to the at least one specified cache server. Identify,
A path that reaches from the guardian to the specified at least one cache server at a predetermined distance is a path that reaches from the guardian to the specified at least one cache server within a predetermined range from the distance of the specified shortest path. A distributed cache system characterized by being. The distributed cache system according to claim 6,
The guardian is
Based on the topology information, identify a plurality of nodes on a path satisfying the specified predetermined condition, and links constituting the path,
Sending subgraph information including information about the identified nodes and the identified links to the cache servers;
The cache server that has received the subgraph information updates the topology information based on the received subgraph information. A cache server selection method for selecting a cache server that transmits data to the client in an apparatus connected to the client and a plurality of cache servers,
The plurality of cache servers construct a network,
Storing topology information of the network;
The method
The device receives a request from the client;
The apparatus specifies at least one cache server that stores data corresponding to the received request from the plurality of cache servers,
For each of the specified at least one cache server, the device has at least one path satisfying a predetermined condition from at least all paths reachable from the device to the specified at least one cache server. Identify based on the topology information,
The apparatus calculates a capacity of each of the specified at least one path for each of the specified at least one cache server;
The apparatus selects a cache server that transmits the data to a client from the identified at least one cache server based on the calculated capacity of each path. The cache server selection method according to claim 11,
The cache server selection method, wherein at least one path satisfying the predetermined condition is a path having a predetermined relationship with the device. The cache server selection method according to claim 12,
A cache server selection method characterized in that a path having a predetermined relationship with the device is a path that reaches a predetermined distance from the device to the specified at least one cache server. The cache server selection method according to claim 13,
The device specifies the shortest path that can be reached from the device to the specified at least one cache server with the shortest distance from all the paths that can be reached from the device to the specified at least one cache server. And
A path that reaches from the apparatus to the specified at least one cache server at a predetermined distance is a path that reaches the specified at least one cache server from the apparatus within a predetermined range from the distance of the specified shortest path. A method for selecting a cache server, comprising: The cache server selection method according to claim 11,
The plurality of cache servers store topology information of the network,
Based on the topology information, the device specifies a node on a path that satisfies the specified predetermined condition, and a link configuring the path,
A cache server selection method comprising: transmitting subgraph information including information relating to the specified plurality of nodes and the specified link to the plurality of cache servers in order to cause the plurality of cache servers to update the topology information. .

Images