JP6139432B2 - Slot group generation device, slot group generation method, and load distribution system - Google Patents

Description

  The present invention relates to a slot group generation device, a slot group generation method, and a load distribution system.

  In a system that accepts service requests over a network, it is necessary to flexibly change the amount of hardware to be allocated according to the amount of service requests in response to an enormous amount of service requests. Realization of has become the mainstream.

  In order to realize a distributed system, a load distribution apparatus that appropriately distributes service requests to one of the servers and equalizes the server load among the servers is required. In many Web services, since each service request can be processed independently, a load balancer that distributes the service requests regardless of the contents of the service requests such as order and randomness is used. However, in a stateful Web service that needs to share the same information among a plurality of service requests such as a SIP (Session Initiation Protocol) application, a load balancer that distributes the related service requests to the same server is provided. It needs to be realized.

  Regarding this necessity, Non-Patent Document 1 applies a consistent hash method in which a data storage server is determined according to a hash value of a key to a load balancer, and the same server as long as the same key exists in a service request. In particular, there is a technique for distributing the load uniformly while suppressing the number of slots in the slot group that defines the association between the hash value and the distribution destination server.

  Non-Patent Document 2 describes a technique for reducing the load bias of the consistent hash method by increasing the number of slots.

Michio Irie et al., “Load Balancing Method Aware of Data Replication in the Consistent Hash Method”, The Institute of Electronics, Information and Communication Engineers, IEICE Technical Report IN2010-77, October 2010 I. Stoica et al., “Chord: A Scalable Peer-to-peer Lookup Service for Internet Applications”, Proc. ACM SIGCOMM 2001, Aug. 2001.

  In the method of Non-Patent Document 1, the number of slot groups can be suppressed, but load equalization is not perfect, and there is a possibility that a large load deviation occurs between servers. The technique of Non-Patent Document 2 also shows that the load bias is reduced by increasing the number of slots, but it is not certain how many slots can achieve the desired uniformity.

  When actually building a system, in a situation where the load is not evenly distributed and it is not clear which server load will be high, select a device with sufficient performance when procuring server hardware. I must. As a result, the procurement cost for system construction will be increased. Therefore, in system construction, equalizing the load among servers in the system is a very important issue.

  Therefore, an object of the present invention is to provide a load distribution system in which there is no load bias among servers.

  In order to solve the above-described problems and achieve the object, the present invention provides a slot group generation device that generates a slot group used for distribution of received packets in a load distribution system that distributes load among a plurality of servers. A slot sequence generation unit that generates a permutation combination of a predetermined number of servers in the load distribution system, a connection graph generation unit that generates a permutation connection graph in which the generated permutation combinations are connected as nodes, and the generated permutation connection graph To generate a slot sequence by arranging a permutation cycle extraction unit for extracting a permutation cycle in which the number of times each node passes is the same among all nodes, and arranging the first servers of the nodes constituting the extracted permutation cycle in the order of the cycles. The slot sequence generation unit and a predetermined hash space are divided into slots having the length of the generated slot sequence, and the division is performed. The slot respectively, characterized in that it comprises a slot group generation unit that generates a slot group allocated to the server in the order indicated in the slot column.

  ADVANTAGE OF THE INVENTION According to this invention, the load distribution system without the bias | biasing of the load between servers can be provided.

FIG. 1 is a configuration diagram illustrating a load distribution system according to each embodiment. FIG. 2 is a configuration diagram illustrating the slot group generation unit. FIG. 3 is a flowchart illustrating a processing procedure of the slot group generation unit according to the first embodiment. FIG. 4 is a flowchart showing the basic slot sequence generation processing of FIG. FIG. 5 is a diagram illustrating an example of generating a basic permutation. FIG. 6 is a diagram illustrating a generation example of a basic permutation connection graph. FIG. 7 is a diagram illustrating an example of generating a basic permutation cycle. FIG. 8 is a diagram illustrating a generation example of a slot string. FIG. 9 is a diagram illustrating a generation example of a slot group in which the assigned sections of the servers A to B are assigned to the hash space. FIG. 10 is a flowchart illustrating a processing procedure of the slot group generation unit according to the second embodiment. FIG. 11 is a flowchart showing the slot string expansion processing of FIG. FIG. 12 is a diagram illustrating an example of generating extended information. FIG. 13 is a diagram illustrating an example of generation of an extended basic permutation. FIG. 14 is a diagram illustrating an example of generating a slot train after expansion. FIG. 15 is a diagram illustrating an example of updating extension information. FIG. 16 is a diagram illustrating a generation example of a slot group in which the assigned sections of the servers A to E are assigned to the hash space. FIG. 17 is a diagram illustrating another configuration example of the load distribution system. FIG. 18 is a diagram illustrating a computer that executes a load distribution program.

  Hereinafter, embodiments of the present invention will be described by being divided into a first embodiment and a second embodiment. In addition, this invention is not limited by each embodiment.

(First embodiment)
As shown in FIG. 1, the load distribution system of each embodiment is configured by connecting a client 100, a load distribution apparatus 200, and a server 300 via a network (not shown).

  The load balancer 200 extracts a predetermined key from the packet reception unit 210 that receives a service request packet from the client 100 and the received packet, and compares the hash value with a slot group to determine a server 300 that is a packet transfer destination. A allocating unit 220 for determining the packet, a packet transferring unit 230 for performing packet transfer based on the determination of the allocating unit 220, a slot group managing unit 240 for generating a slot group for realizing uniform load, and a slot group A slot group generalization unit 250 that controls abstraction of server names and the like so that the management unit 240 can be used by many load distribution devices.

  The distribution unit 220 extracts a key extraction unit 221 from the packet received by the packet reception unit 210 as a key by the consistent hash method, generates a hash value from the extracted key, and generates a slot group generalization unit A packet transmission destination determination unit 222 that searches a slot corresponding to the hash value from the slot group acquired from 250 and determines a server assigned to the slot as a packet transfer destination.

  The slot group management unit 240 receives various parameters such as fault tolerance indicating the server list and the number of servers that have failed until the load is leveled, and generates a slot group. Group generation device) 241 and a slot group holding unit 242 for holding the slot group generated by the slot group generation unit 241.

  The slot group generalization unit 250 assigns a general server name to each server based on the load distribution system server list set in advance by the system maintainer 400 or the like, and sends the slot group management unit 240 the slot group information. Server group abstraction unit 251 that requests generation, and server group individualization unit that converts a server name assigned to each slot from a generic name to a system-specific name in the slot group generated by slot group management unit 240 252 and a server list holding table 253 that holds a correspondence table related to generalization of the server list. 1 is not limited to the number illustrated in FIG. 1, and may be an arbitrary number for the set of servers 300.

  In the following description, a case where the application service of the load distribution apparatus 200 is a telephone service that requires session processing will be described as an example, but the application service is not limited to this.

  The client 100 is, for example, a SIP (Session Initiation Protocol) client such as a telephone terminal, and is a device that transmits a request packet related to session management.

  In this case, the load balancer 200 and the set of servers 300 behave as if they are SIP servers as an integrated system. The SIP packets from the client 100 are exclusively received and distributed by the load balancer 200, and the session control itself is the server 300. Implement. Thus, the client 100 may face the load distribution apparatus 200 without being aware of the addition of the server 300 or the like. However, since session processing related to the same client 100 needs to be performed consistently by the same server 300, the distribution logic of the load distribution apparatus 200 needs to distribute SIP packets from the same client 100 to the same server 300. There is. In this regard, the load balancer 200 uses a consistent hash method described below.

  In the consistent hash method, the packet reception unit 210 of the load distribution apparatus 200 that has received the SIP packet from the client 100 passes the received SIP packet to the key extraction unit 221 of the distribution unit 220. In order to identify the client 100, the key extraction unit 221 extracts client identification information such as From and Call-ID from the SIP header and sends it to the packet transmission destination determination unit 222. The packet transmission destination determination unit 222 that has received the client identification information generates a hash value of the client identification information, and acquires the hash value of the hash value handled by each server 300 from the slot group generalization unit 250. By comparing the hash value of the client identification information, the server 300 as the transfer destination of the received packet is determined. By determining the distribution destination in this way, packets can be transferred to the same server 300 as long as the client identification information is the same.

  After the packet transfer destination is determined, the packet transfer unit 230 rewrites the packet destination IP address and the like based on the information, and transfers the packet to the corresponding server 300.

  When the load distribution apparatus 200 distributes the packets by the above processing, in order to equalize the load among the servers 300, a slot group in which the range of hash values that each server 300 is in charge of is generated is generated. There is a need to. Hereinafter, a method of generating a slot group in which the range of hash values that each server 300 is in charge of, including the case where several servers 300 have failed, will be described.

  In this example, it is assumed that there are five servers 300 constituting the load balancing system, and the server names are SIPserver1, SIPserver2, SIPserver3, SIPserver4, and SIPserver5, respectively. In addition, regarding the response when the server 300 fails, the load is made uniform on the remaining server 300 even if one of the servers 300 randomly fails.

  The slot group generalization unit 250 is responsible for name conversion so that the slot group management unit 240 can generate a slot group without being aware of the individual server name SIPserver1 or the like.

  Specifically, in the server list holding table 253, the system maintainer 400 associates a server name unique to the system such as SIPserver1 in advance with a server name used by the slot group management unit 240 such as A and B, for example, SIPserver1: A, SIPserver2: B, SIPserver3: C, SIPserver4: D, SIPserver5: E are held. In this situation, when the server group abstraction unit 251 receives a request from the system maintenance person 400 to generate a slot group that is resistant to one of the five servers, the server group abstraction unit 251 maintains the server list. A request is made to the slot group generation unit 241 of the slot group management unit 240 to generate slot groups related to A, B, C, D, and E based on the information in the table 253. In this case, for example, “the number of required failure tolerances (the number of assumed failure servers) +1”, that is, “1 + 1 = 2” here is transmitted as a parameter relating to slot group generation (hereinafter, this parameter is referred to as “parameter”). Called the basic permutation length). Note that the process of adding “1” is not necessarily performed by the server group abstraction unit 251 and may be performed by the slot group generation unit 241.

  In this case, for example, when the server group abstraction unit 251 transmits information of the server list “A, B, C, D, E” and the basic permutation length “2” to the slot group generation unit 241, From the management unit 240, for example, a slot group as shown in FIG. That is, a slot group to which the assigned range of each server is assigned is returned to the circular hash space. Then, the server group individualization unit 252 converts the returned slot group from a general name such as A or B to an individual name such as SIPserver1 based on the information in the server list holding table 253. Thus, a slot group corresponding to an individual name such as SIPserver1 is generated.

  This slot group is also used when a server that takes over the processing of the failed server is selected when any server in the load balancing system fails. For example, after SIP server 1 has been selected as the distribution destination of the received packet by distribution unit 220, when SIP server 1 fails, the server to which the process is taken over is the right-hand side of the failed server (SIP server 1 (A)) in the relevant slot group ( A server (for example, SIPserver5 (E)) arranged in the clockwise direction is selected.

  Next, the slot group generation unit 241 will be described in detail. As shown in FIG. 2, the slot group generation unit 241 includes a slot sequence generation unit 243, a connection graph generation unit 244, a permutation cycle extraction unit 245, a slot sequence generation unit 246, and a slot allocation unit 247. . The extended basic permutation generation unit 248 and the expansion slot sequence generation unit 249 may or may not be equipped with the slot group generation unit 241. The case of being equipped will be described in the second embodiment. Details of the processing of each unit will be described later with specific examples.

  The slot string generation unit 243 generates a permutation combination of servers of (the number of assumed failure servers + 1) from the load balancing system.

  The connection graph generation unit 244 generates a permutation connection graph in which the permutation combinations generated by the slot sequence generation unit 243 are connected as nodes.

  The permutation cycle extraction unit 245 extracts a permutation cycle that passes through each node the same number of times from the permutation connection graph generated by the connection graph generation unit 244.

  The slot sequence generation unit 246 generates a slot sequence (basic slot sequence) by arranging the leading servers of the nodes constituting the permutation cycle extracted by the permutation cycle extraction unit 245 in the order of the cycles.

  The slot allocation unit 247 generates a slot group using the slot sequence generated by the slot sequence generation unit 246. Specifically, the slot allocation unit 247 divides the hash space used for distribution of received packets into slots having the length of the slot sequence generated by the slot sequence generation unit 246. Then, the slot allocation unit 247 generates a slot group in which servers are allocated to each of the divided slots in the order shown in the slot row.

  Next, a processing procedure of the slot group generation unit 241 will be described.

  As shown in FIG. 3, the slot group generation unit 241 obtains a server list indicating the servers that are the target of slot group generation (S10), and generates a basic slot sequence of the servers indicated in the server list (S20). Then, a slot group is generated based on the generated slot string (S30).

  For example, when A, B, and C are described in the server list and the slot group generation unit 241 generates a slot string called ABCBAC by the basic slot string generation processing, a slot group is generated based on the slot string called ABCBAC. For example, as shown in FIG. 9 to be described later, the slot group generation unit 241 divides the hash space into six equal parts (equally divided by the length of the slot string), and assigns servers to the slots in the same order as the slot string. That is, the slot group generation unit 241 generates a slot group of A → B → C → B → A → C (→ returns to A). As described above, the slot group generation unit 241 generates a slot group in which the number of appearances of each server is the same and the number of times other servers appear next to each server is uniform. By distributing received packets according to the slot group, the load can be made uniform among the servers, and the load at the time of failure can also be made uniform.

  The basic slot sequence generation process (S20) in FIG. 3 is performed according to the procedure shown in FIG. First, the slot sequence generation unit 243 of the slot group generation unit 241 generates a basic permutation based on the server list acquired in S10 of FIG. 3 (S21). Specifically, the slot sequence generation unit 243 extracts all permutation combinations of servers having the number of basic permutation lengths (the number of assumed failure servers + 1), that is, the server group indicated in the server list.

For example, let us consider a case where the assumed number of failed servers is “1” and A, B, and C are described in the server list. In this case, as shown in FIG. 5, the slot sequence generation unit 243 enumerates all permutations having a length of “2 (basic permutation length)” for the three servers. AB, AC, BC, BA, CA, CB It becomes. In other words, for example, when A fails, B takes over processing, when A fails, C takes over, when B fails, C takes over processing, and so on. List all permutations with servers that take over The permutations enumerated in this way are defined as basic permutations. The number of basic permutations when there are three servers is ₃ P ₂ = 3! That is, it becomes six pieces.

  Returning to the description of FIG. After S21, the connection graph generation unit 244 generates a basic permutation connection graph (S22). Specifically, the connection graph generation unit 244 generates a basic permutation connection graph in which the permutation combinations (basic permutations) extracted in S1 are connected as nodes.

  For example, as illustrated in FIG. 6, when the basic permutation is AB, AC, BC, BA, CA, CB, the connection graph generation unit 244 has a length of 2-1 = 1 for each basic permutation. Connect to a basic permutation starting with a permutation. That is, the connection graph generation unit 244 generates a node having each basic permutation as a label, and extracts a partial permutation excluding the first one server for each basic permutation. For example, a node having a basic permutation of AB, AC, BC, BA, CA, CB as a label is generated, a partial permutation of B is extracted for AB, and a partial permutation of C is extracted for AC. Such processing is performed for all basic permutations. And the connection graph production | generation part 244 extracts the basic permutation which starts from the extracted partial permutation. For example, BC and BA are extracted as basic permutations starting from a partial permutation called B, and CA and CB are extracted as basic permutations starting from a partial permutation called C (see “Link of Basic Permutation Connection Graph” in FIG. 6). As described above, the connection graph generation unit 244 extracts a basic permutation having the same server as the end of the basic permutation as the head of the graph as a link destination of the graph. Then, the connection graph generation unit 244 connects the nodes indicating the basic permutations with the link destination nodes extracted by the above-described processing through the directed links, and generates a basic permutation connection graph as illustrated on the right side of FIG. .

  Returning to the description of FIG. After S22, the permutation cycle extraction unit 245 extracts a basic permutation cycle that passes through each node the same number of times from the basic permutation connection graph generated in S22 (S23).

  For example, the permutation cycle extraction unit 245 extracts a cycle that passes through each node the same number of times from a basic permutation connection graph as shown in FIG. 7, and obtains a basic permutation cycle shown on the right side of FIG. If such a closed path is generated, the slot group generation unit 241 may determine the order of the servers on the slot string along the closed path, thereby generating a slot string in which each basic permutation appears the same number of times. it can. There is no limitation on the method of extracting the cycle, but at least the graph satisfies the condition having the Euler cycle, and therefore the Euler cycle can be extracted by using Fleury's algorithm.

  In the basic permutation connection graph illustrated in FIG. 7, each node is a second-order directed graph for both the incoming order and the outgoing order. Therefore, the extracted Euler circuit is a circuit that passes through each node exactly twice. In order to deal with a more general server list or the like, a method of extracting the Euler cycle is assumed. However, in this example, it is clear that a cycle that passes through each node once can be extracted. Then, the basic permutation circuit as shown in FIG. 7 uses “AB → BC → CB → BA → AC → CA → (return to AB)”.

  Returning to the description of FIG. After S23, the slot sequence generation unit 246 generates a slot sequence by arranging the leading servers of the nodes constituting the basic permutation cycle extracted in S23 in the order of the cycle (S24). For example, as shown in FIG. 8, the slot sequence generation unit 246 scans the basic permutation cycle in the order of AB → BC → CB → BA → AC → CA, and arranges the servers at the beginning of the respective labels along the cycle. ABCBAC is generated as a column.

  Thereafter, for example, as shown in FIG. 9, the slot allocation unit 247 equally divides the hash space by the length of the slot string generated in S24 of FIG. 4 (divides the space into six equal parts), and the slot string (ABCBAC) A slot group in which servers are assigned to each slot in the same order as is generated. That is, the slot allocation unit 247 generates a slot group of A → B → C → B → A → C (→ returns to A).

  In the slot group generated in this way, all the basic permutations described above are included the same number of times. Therefore, by using such a slot group for distribution of received packets, the distribution unit 220 can equalize the load among the servers including when a server in the load distribution system fails. That is, since the same number of servers and server permutations appear in the slot group described above, the number of slots assigned in normal times is uniform. Further, for example, when there is a part assigned in the order of A → B as slot assignment, if A follows a consistent hash method when a failure occurs, the next B is assigned. As a result, all basic permutations Are the same, the number of times other servers appear next to each server is uniform, and the load can be made uniform even at the time of failure.

(Second Embodiment)
Next, a second embodiment of the present invention will be described. In the second embodiment, the slot group generation unit 241a generates a slot sequence of some servers in the server list (basic slot sequence generation), and extends the slot sequence to the remaining servers in the server list (slots). Column expansion). As described above, the slot group generation unit 241a generates the slot string by combining the basic slot string generation and the slot string expansion, so that even when the number of servers indicated in the server list is large, the slot string of all servers can be generated. Such calculation amount and slot length can be reduced.

  Hereinafter, the slot group generation unit 241a first generates a slot sequence of (basic permutation length + 1) servers among the servers in the server list, and then expands the slot sequence to the remaining servers. Explained as an example. The same configurations as those of the first embodiment described above are denoted by the same reference numerals and description thereof is omitted.

  As shown in FIG. 2, the slot group generation unit 241a further includes an extended basic permutation generation unit 248 and an expansion slot sequence generation unit 249 indicated by broken lines.

  First, the slot group generation unit 241a uses the slot sequence generation unit 243, the connection graph generation unit 244, the permutation cycle extraction unit 245, and the slot sequence generation unit 246 to determine the servers (for example, A to E) indicated in the server list. Among them, a slot sequence of (basic permutation length + 1) servers, for example, three servers A, B, and C is generated.

  The extended basic permutation generation unit 248 is an extension that is a permutation combination of (basic permutation length-1) servers from the servers (for example, A, B, and C) indicated in the basic permutation generated by the slot sequence generation unit 243. Extract all basic permutations. For example, the extended basic permutation generation unit 248 extracts A, B, and C as the extended basic permutation.

  The expansion slot string generation unit 249 selects one of the servers (for example, A to E) indicated in the server list, which is not targeted by the slot string generation unit 246 as a slot string (basic slot string).

  For example, the expansion slot sequence generation unit 249 selects either D or E, and the expansion slot sequence generation unit 249 uses the expansion basis from the slot sequence (for example, ABCBAC) generated by the slot sequence generation unit 246. One server (for example, A, B, C) shown in the permutation is extracted one by one, and each expansion target server (for example, D) is replaced with a permutation sandwiched between expansion basic permutations (for example, A, B, C). . As a result, a new slot sequence in which the expansion target server is inserted is generated in the basic slot sequence. That is, the basic slot sequence is expanded to the expansion target server. Thereafter, the expansion slot string generation unit 249 inserts an expansion target server (for example, E) further into the slot string in which the expansion server (for example, D) is inserted, by the same processing as described above. As a result, a slot train of all servers (A, B, C, D, E) shown in the server list is generated.

  Details of the extended basic permutation generation unit 248 and the expansion slot sequence generation unit 249 will be described later using a specific example.

  Next, a slot group generation processing procedure in the slot group generation unit 241a of the second embodiment will be described.

  As shown in FIG. 10, the slot group generation unit 241a acquires basic information indicating the server list and the basic permutation length (S40), and extracts a server list for basic slot sequence generation based on the basic information. (S50). Then, the slot group generation unit 241a generates a basic slot sequence based on the server list extracted in S50 by the same procedure as S20 in FIG. 3 (S20). Then, the slot group generation unit 241a generates extended information (S60). This extension information is information used to extend the slot sequence, and includes a server list for generating a basic slot sequence (pre-expansion server list), a basic slot sequence (slot sequence before expansion) generated in S20, This is information indicating the basic permutation length and a list of servers (extended server list) excluded from the basic slot sequence in the server list.

  For example, as shown in FIG. 12, when the server list indicated in the basic information is “A, B, C, D, E” and the basic permutation length is “2”, the slot group generation unit 241a Permutation length + 1 = 3 servers are arbitrarily extracted, and a server list (for example, A, B, C) for basic slot sequence generation is extracted. Here, servers that are not subject to extraction (for example, D and E) are put on the extended server list. Thereafter, the slot group generation unit 241a generates a slot sequence by a basic slot sequence generation process (a process similar to S20 in FIG. 3). This slot row is a basic slot row. Then, in the slot group generation unit 241a, the pre-expansion server list (server list for basic slot sequence generation) is “A, B, C”, and the slot sequence before expansion (basic slot sequence) is “ABCBAC”. The extended information indicating that the basic permutation length is “2” and the extended server list is “D, E” is generated.

  Returning to the description of FIG. After S60, the slot group generation unit 241a executes a slot string expansion process based on the extension information generated in S60 (S70), and uses the expansion slot string generated in S70 to perform S30 in FIG. A slot group is generated by the same processing (S30).

  Here, the slot string expansion process in S70 of FIG. 10 will be described with reference to FIG. When the extended basic permutation generation unit 248 acquires the extended information generated in S60 (S71), the extended basic permutation generation unit 248 extracts an expansion target server (S72) and generates an extended basic permutation (S73).

For example, the extended basic permutation generation unit 248 extracts D from the extended server list “D, E” of the extended information shown in FIG. Then, as shown in FIG. 13, the extended basic permutation generation unit 248 uses three servers A, B, and C from the information that the pre-extension server list is “A, B, C” and the basic permutation length is “2”. For all permutations with a basic permutation length 2-1 = 1. The permutation extracted here is called an extended basic permutation. For example, the extended basic permutation generation unit 248 generates “A, B, C” as the extended basic permutation. The number of extended basic permutations extracted here is ₃ P ₁ = 3 when there are three servers A, B, and C.

  Returning to the description of FIG. After S73, the expansion slot string generation unit 249 inserts the expansion basic permutation generated in S73 into the slot string (S74).

For example, the expansion slot string generation unit 249 searches for the position of the expansion basic permutation (for example, A, B, C) generated in S73 one by one from the slot string (for example, ABCBAC) before the expansion. Go. The number of times of searching is determined by dividing the slot sequence length before expansion by the number of basic permutations, which is equal to the number of times each node has been passed in the basic permutation cycle of the slot sequence (see FIG. 7). . For example, in the example described above, the length of the slot sequence (ABCBAC) before expansion is “6” and the number of basic permutations is “ ₃ P ₂ = 6”, so “6 ÷ 6 = 1”.

  Then, the expansion slot string generation unit 249 replaces each expansion basic permutation searched from the slot string before expansion with a permutation sandwiching the expansion target server in the expansion basic permutation. For example, consider a case where the expansion target server is “D”. In this case, as illustrated in FIG. 14, the expansion slot sequence generation unit 249 replaces the extension basic permutation “A” with “ADA” for one location that is “A” in the slot sequence before expansion. To do. Similarly, one location “B” in the slot row before expansion is replaced with “BDB”. Further, “CDC” is substituted for one location that is “C” in the slot row before expansion. As a result, “ADABCBDDBACDC” is obtained as the expanded slot string.

  In this way, the expansion slot sequence generation unit 249 extracts one each of the expansion basic permutation from the slot sequence before expansion, and replaces each expansion target server with a permutation sandwiched between the expansion basic permutations. In the above example, the case where the length of the extended basic permutation is “1” has been described. However, when the length is “2”, for example, when the extended basic permutation is “AB”, the extension target If the server is “D”, the place “AB” in the slot string before expansion is replaced with “ABDAB”.

  Returning to the description of FIG. After inserting the expansion basic permutation into the slot string in this way (S74), the expansion slot string generating unit 249 updates the extension information by adding the processed expansion target server to the server list before expansion (S75). . For example, after inserting the server “D” into the slot string “ABCBAC” before expansion, the expansion slot string generation unit 249 updates the left extension information in FIG. 15 to the right extension information.

  Returning to the description of FIG. If the extension target server remains in S76 (Yes in S76), the slot group generation unit 241a executes the processing subsequent to S71 again using the updated extension information. On the other hand, if the expansion target server does not remain (No in S76), the slot string expansion process is terminated, and the process proceeds to S30 in FIG.

  The expansion slot string generation unit 249 inserts the expansion target server “D” into the slot string by the process shown in FIG. 14, and then inserts the expansion target server “E” into the slot string by the same process as described above. That is, the expansion slot string generation unit 249 generates a slot string “ADABCBDDBACDC”, and then extracts one place where “A, B, C, D” of this slot string appears one by one. Replacement is performed with a permutation sandwiching the server “E”. As a result, the expansion slot string generation unit 249 generates a slot string “AEADABEBBCBDDBDACDCEC”.

  When the slot sequence incorporating all the servers (A, B, C, D, E) shown in the server list is completed by the above processing, the slot allocation unit 247 of the slot group generation unit 241a uses the slot sequence based on this. Create a group.

  For example, the slot group allocation unit 247 of the slot group generation unit 241a generates a slot obtained by dividing the hash value space into 20 slots of the slot sequence “AEADABEBCBBDDBACDEC” as illustrated in FIG. The server in charge is assigned in the order of appearance in the column “AEADABEBBCBDDEDBACDEC”.

  As a result of the above processing, a slot group including all the servers shown in the server list and having a uniform assigned range among them is generated. Therefore, in the load balancing system using the slot group, the load is distributed between the servers. It becomes possible to disperse uniformly.

  In addition, the slot group generation unit 241a generates up to (basic permutation length + 1) slots in the slot train, and inserts the servers that exceed (basic permutation length + 1) into the slot train by the slot train expansion process. Thus, it is possible to reduce the calculation amount and the slot length when generating the slot train for all the servers shown in the server list.

(Other embodiments)
The slot group management unit 240 in FIG. 1 is illustrated as being one in the load balancer 200, but is cut out from the load balancer 200 as shown in FIG. In this case, one slot group providing apparatus 500 can perform processing of a plurality of load distribution apparatuses 200.

(program)
It is also possible to create a program in which the processing executed by the load distribution apparatus 200 according to the above embodiment is described in a language that can be executed by a computer. In this case, the same effect as the above-described embodiment can be obtained by the computer executing the program. Further, such a program may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read by a computer and executed to execute the same processing as in the above embodiment. Hereinafter, an example of a computer that executes a load distribution program that implements the same unit as the load distribution apparatus 200 will be described.

  FIG. 18 is a diagram illustrating a computer that executes a load distribution program. As illustrated in FIG. 18, the computer 1000 includes, for example, a memory 1010, a CPU 1020, a hard disk drive interface 1030, a disk drive interface 1040, a serial port interface 1050, a video adapter 1060, and a network interface 1070. These units are connected by a bus 1080.

  The memory 1010 includes a ROM (Read Only Memory) 1011 and a RAM 1012. The ROM 1011 stores a boot program such as BIOS (Basic Input Output System). The hard disk drive interface 1030 is connected to the hard disk drive 1090. The disk drive interface 1040 is connected to the disk drive 1100. A removable storage medium such as a magnetic disk or an optical disk is inserted into the disk drive 1100, for example. For example, a mouse 1110 and a keyboard 1120 are connected to the serial port interface 1050. For example, a display 1130 is connected to the video adapter 1060.

  Here, as shown in FIG. 18, the hard disk drive 1090 stores, for example, an OS 1091, an application program 1092, a program module 1093, and program data 1094. Each table described in the above embodiment is stored in the hard disk drive 1090 or the memory 1010, for example.

  Further, the load distribution program is stored in the hard disk drive 1090 as a program module in which a command executed by the computer 1000 is described, for example. Specifically, a program module describing each process executed by the load distribution apparatus 200 described in the above embodiment is stored in the hard disk drive 1090.

  Further, data used for information processing by the load balancing program is stored as program data in, for example, the hard disk drive 1090. Then, the CPU 1020 reads out the program module 1093 and the program data 1094 stored in the hard disk drive 1090 to the RAM 1012 as necessary, and executes the above-described procedures.

  Note that the program module 1093 and the program data 1094 related to the load balancing program are not limited to being stored in the hard disk drive 1090. For example, the program module 1093 and the program data 1094 are stored in a removable storage medium and read by the CPU 1020 via the disk drive 1100 or the like. May be issued. Alternatively, the program module 1093 and the program data 1094 related to the load distribution program are stored in another computer connected via a network such as a LAN (Local Area Network) or a WAN (Wide Area Network), and are transmitted via the network interface 1070. May be read by the CPU 1020.

DESCRIPTION OF SYMBOLS 200 Load distribution apparatus 210 Packet receiving part 220 Distribution part 221 Key extraction part 222 Packet transmission destination determination part 230 Packet transfer part 240 Slot group management part 241, 241a Slot group generation part 242 Slot group holding part 243 Slot row generation part 244 Connection Graph generation unit 245 Permutation cycle extraction unit 246 Slot sequence generation unit 247 Slot allocation unit 248 Extended basic permutation generation unit 249 Expansion slot sequence generation unit 250 Slot group generalization unit 251 Server group abstraction unit 252 Server group individualization unit 253 Server list Holding table

Claims (5)

In a load distribution system that distributes load among a plurality of servers, a slot group generation device that generates a slot group used for distributing received packets,
A first slot sequence generation unit that generates a permutation combination of servers having a basic permutation length obtained by adding 1 to the number of one or more failed servers assumed in the load distribution system from the servers constituting the load distribution system ;
A connection graph generation unit for generating a permutation connection graph in which the generated permutation combinations are connected as nodes;
From the generated permutation connection graph, a permutation cycle extraction unit that extracts a permutation cycle in which the number of times each node passes is the same between all nodes;
A second slot string generation unit that generates a slot string by arranging the leading servers of the nodes constituting the extracted permutation cycle in the order of the cycles;
A slot group generation unit that divides a predetermined hash space into slots having a length of the generated slot string, and generates a slot group in which the servers are assigned to each of the divided slots in the order indicated by the slot string; A slot group generation device comprising: The slot group generation device includes:
When a server that is excluded from the slot column is incorporated into the slot column, an extended basic permutation that is a permutation combination of the number of servers obtained by subtracting 1 from the basic permutation length is generated for the server indicated in the slot column. An extended basic permutation generator;
An expansion slot sequence that generates a permutation in which the servers excluded from the slot sequence are sandwiched by the extended basic permutation, and generates a slot sequence that replaces the location of the extended basic permutation in the slot sequence with the generated permutation The slot group generation device according to claim 1 , further comprising a generation unit. In a load distribution system that distributes load among a plurality of servers, a slot group generation device that generates a slot group used for distribution of received packets includes:
Generating a permutation combination of servers having a basic permutation length obtained by adding 1 to the number of one or more failed servers assumed in the load distribution system from the servers constituting the load distribution system ;
Generating a permutation connection graph connecting the generated permutation combinations as nodes;
Extracting a permutation cycle in which the number of passes of each node is the same between all nodes from the generated permutation connection graph;
Arranging the first servers of the nodes constituting the extracted permutation cycle in the order of the cycles to generate a slot sequence;
Performing a step of dividing a predetermined hash space into slots having the length of the generated slot sequence, and generating a slot group in which the servers are assigned to the divided slots in the order indicated by the slot sequence. A slot group generation method characterized by the above. The slot group generation device comprises:
When a server that is excluded from the slot column is incorporated into the slot column, an extended basic permutation that is a permutation combination of the number of servers obtained by subtracting 1 from the basic permutation length is generated for the server indicated in the slot column. Steps,
Generating a permutation in which the servers excluded from the slot sequence are sandwiched by the extended basic permutation, and generating a slot sequence by replacing the location of the extended basic permutation in the slot sequence with the generated permutation. The slot group generation method according to claim 3 , further executing. In a load distribution system that distributes load among a plurality of servers, a slot group generation device that generates a slot group used for distribution of received packets, and distribution that distributes the received packets using the generated slot groups A load balancing system comprising a device,
The slot group generation device includes:
A first slot sequence generation unit that generates a permutation combination of servers having a basic permutation length obtained by adding 1 to the number of one or more failed servers assumed in the load distribution system from the servers constituting the load distribution system ;
A connection graph generation unit for generating a permutation connection graph in which the generated permutation combinations are connected as nodes;
From the generated permutation connection graph, a permutation cycle extraction unit that extracts a permutation cycle in which the number of times each node passes is the same between all nodes;
A second slot string generation unit that generates a slot string by arranging the leading servers of the nodes constituting the extracted permutation cycle in the order of the cycles;
A slot group generation unit that divides a predetermined hash space into slots having a length of the generated slot string, and generates a slot group in which the servers are assigned to each of the divided slots in the order indicated by the slot string; A load distribution system comprising:

Images