JP4806774B2 - Network system, information processing apparatus, information transfer method, program, and recording medium - Google Patents

Description

  The present invention relates to a network system, an information processing apparatus, an information transfer method, a program, and a recording medium, and more particularly to a network system in which a plurality of information processing apparatuses are connected and communication between information processing apparatuses is performed by a packet transfer method.

  With regard to network management of peer machine-to-platform peer-to-peer communications and access to distributed files performed by packet transfer (bucket relay system) via adjacent platforms, the existing technologies can be broadly divided into two.

  One is a method in which a search message is propagated between adjacent nodes without particularly having a structure in the network (see Flooding Search, Non-Patent Document 1, Non-Patent Document 2).

  The other is to give a structure to the network and perform search, route maintenance, and information transfer based on distributed hash table management based on the structure (for example, Chord's method (See Patent Document 3, see Non-Patent Document 4 for Pastry method, and Non-Patent Document 5 for Tapestry method.)

Gnutella 2003, The Gnutella web site: http://gnutella.wego.com. Ripeanu, M. and Foster, I. 2002, Mapping the Gnutellanetwork: Macroscopic properties of large-scale peer-to-peer systems, InProceedings of the 1st International Workshop on Peer-to-PeerSystems (IPTPS'02). Stoica, I. et.al, 2001, Chord: A scalablepeer-to-peer look-up service for internet application, In Proceedings of SIGCOMM 2001. Rowstron, A and Druschell, P. 2001, Pastry: Scalable distributed object location and routing for large-scale peer-to-peersystems, In proceeding of IFIP / ACM Middleware, Heidelberg, Germany. Zhao, B. et.al, 2001, Tapestry: Aninfrastructure for fault-torelant wide-area location and routing, Tech. Rep.UCB / CSD-01-1141, Computer Science Division, University California, Berkerey.

  Typical network management costs include the cost of route maintenance (such as destination search costs, route maintenance costs, etc., including information transfer costs) due to platform machine detachment, etc., and the connection of new entrants. Cost (insertion location search cost, insertion cost, post-insertion path configuration cost, etc.).

  However, in Flooding Search, the search message is randomly distributed throughout the network. For this reason, the cost required for route reconfiguration and the node search cost are proportional to the number of network nodes in the worst case, and the route length is also proportional to the number of nodes in the worst case. In addition, the amount of communication increases, which is not suitable for large scale.

  In the Chord method using a tree network, for example, it is necessary to repair the number of adjacent nodes depending on the height of the worst tree when a new node enters. This complicates the processing, and the repair message propagates between the nodes, increasing the amount of communication and increasing the load on the network. The Pastry method is also the same as the tree structuring, and complicated processing is required for network management, resulting in an increase in communication volume due to message propagation. The same applies to the Tapestry method.

  In view of this, the present invention provides a network that allows a communication path length between arbitrary information processing units to be kept small in a dynamically changing network, and further allows the repair and maintenance of the communication path to be performed at low cost. The purpose is to provide a system.

  The invention according to claim 1 is a network system in which a plurality of information processing devices are connected and communication between the information processing devices is performed by a packet transfer method, and each of the information processing devices manages one or a plurality of communication bases. One or more management means are provided, each management means has a communication path search processing means for making a search request for communication paths to other management means, and each management means is identified with respect to other management means And at least one of the management means having the same calculated value calculated by a predetermined calculation formula based on the identification information is a node management means, and the other management means is an annular chamber management means The node management means and the ring tuft management means of the same calculated value are connected in a ring, and the node management means is connected to one or more of the node management means of different calculated values, and the communication path search of the node management means is performed. The processing means requests the establishment of the communication path from the managed communication base to the communication base managed by the other management means, and when the communication path to the other management means is not established, the other management Depending on the information held about the means, a communication path search request is made to the annular chamber management means having the same calculated value, or a communication path search is made to part or all of the section management means having different calculated values to be connected. It is a network system that makes a request or makes a search request for a communication path to the annular cell management means and the node management means of different calculated values to be connected.

  The invention according to claim 2 is the network system according to claim 1, wherein each of the clause management means is one in the management means of the same calculation value, and the clause management means performs a predetermined calculation based on the identification information. An order tree based on a calculated value calculated by an expression is configured, and the communication path search processing unit of the clause management unit holds the identification information of the other management unit based on the identification information of the other management unit. If the calculated value calculated by the predetermined calculation formula is the same as the calculated value calculated by the predetermined calculation formula based on its own identification information, a request for searching the communication path is made to the annular chamber management means, and the calculation value is different. In some cases, a search request for a communication path is made to some or all of the section management means for different calculated values connected according to the configuration of the ordered tree.

  The invention according to claim 3 is the network system according to claim 2, wherein the management means is in a circular connection relationship with the management means to be disconnected when the other management means to be connected is disconnected from the connection relation. When the disconnecting management means and the other management means in the annular connection relationship are caused to form an annular connection relationship, and the leaving management means is a node management means having an annular tuft management means When one of the annular tuft management means has a connection relationship with the position on the order tree of the leaving node management means, and when it does not have, the management means located in the descendant of the order tree of the leaving management means Of these, there is a connection relationship management means for generating a connection relationship in which one of the values close to the calculated value of the leaving management means is a position on the ordered tree of the leaving management means.

  As in the inventions according to claim 2 and claim 3, the clause management means constitutes an ordered tree, so that, for example, it is possible to narrow down the search space, and when a new management means is connected or disconnected. It is possible to design a system suitable for operation so that the influence is locally stopped and a new node can be connected with O (log n) for the number of nodes n.

  The invention according to claim 4 is the network system according to claim 2 or 3, wherein the management unit has a connection relationship with another management unit in addition to the connection relationship forming the ordered tree with a ring tuft. The communication path search processing means also makes a search request for a communication path to a management means having a connection relationship other than the connection relationship that constitutes the ordered tree with a ring tuft, and the management means issues a search request for the communication path. If received, it is determined whether or not a request for searching for a communication path has been received redundantly, and a duplicate search eliminating means for performing a search process for a communication path only when there is no duplication is provided.

  In the search on the ordered tree with ring tufts, the ratio of the leaf nodes to the total nodes is 50% from the structure of the ordered tree, and 75% when including the parent nodes of the leaf nodes. According to the invention according to claim 4, the communication path search processing means also uses connection relations other than the connection relations that constitute the circular tufted ordered tree, so that the search target node close to the leaf node from the node node close to the leaf node The possibility of forming a route to is increased, and the route length can be shortened. Then, the redundant search exclusion means excludes the same search, thereby preventing a useless search due to multiplexing of search paths.

  In the invention according to claim 4, the transfer of the search request for the communication path is not limited in the case of connection relations constituting the order tree, and in the case of connection relations other than the connection relation constituting the order tree. The number of transfers may be limited.

The invention according to claim 5 is the network system according to any one of claims 1 to 4, wherein there is a precious relationship between the communication bases , and the communication path search processing means is a communication path establishment target. When it is not possible to reach the management means for managing the communication mother and the management means for managing the communication mother that is in the relationship of the communication mother for which the communication path is to be established, The communication path search request is made to the management means for managing the communication base in the network.

  According to the invention according to claim 5, it is possible to realize a search for a communication path based on the lowest community among the upper communities of the communication base to be searched, and it is expected to reduce the search space. It becomes possible.

  The invention according to claim 6 is an information processing apparatus in which communication with another information processing apparatus is performed by a packet transfer method, and includes one or a plurality of communication management means, and each of the communication management means has another communication Communication path search processing means for making a search request for a communication path to the management means, and each communication management means is provided with identification information enabling identification with respect to other communication management means, and the identification information includes The communication management means having the same calculation value calculated based on a predetermined calculation formula based on it are directly or indirectly connected via the communication management means having the same calculation value, and at least one of the communication management means having the same calculation value is different. When there is a direct connection relationship with the calculated value communication management means, the communication management means is requested to establish a communication path to another communication management means, and a communication path to the other communication management means is not established. In addition, according to the information held regarding the other communication management means, a communication path search request is made to the communication management means with the same calculated value, or a part or all of the communication management means with different calculated values It is an information processing apparatus that makes a search request for a communication path or makes a search request for a communication path to a communication management unit having the same calculated value and a communication management unit having a different calculated value.

  The invention according to claim 7 is an information transfer method in an information processing apparatus in which communication with another information processing apparatus is performed by a packet transfer method, and the information processing apparatus includes one or a plurality of communication management units, Each of the communication management means has a communication path search processing means for making a search request for a communication path to other communication management means, and each of the communication management means enables identification with respect to other communication management means Communication management means having the same calculation value calculated by a predetermined calculation formula based on the identification information is directly connected or indirectly connected via the communication management means of the same calculation value, and the same calculation value At least one of the communication management means is directly connected to the communication management means of different calculated values, and the communication management means is requested to establish a communication path to the other communication management means, and the other communication management means When a communication route for the communication control unit is not established, a communication route search request is issued to the communication management unit having the same calculated value or communication management of a different calculated value is performed according to the information held regarding the other communication management unit. Information including a step of making a communication path search request to a part or all of the means, or making a communication path search request to a communication management means having the same calculated value and a communication management means having a different calculated value It is a transfer method.

  The invention according to claim 8 is a program for causing a computer to realize the information transfer method according to claim 7.

  The invention according to claim 9 is a recording medium on which the program according to claim 8 is recorded.

  In the first aspect of the invention, the predetermined calculation formula may be a hash function, and the same hash value management means may be distinguishable by a hash value calculated by another hash function. By using such a binary hash value and the structure of an ordered tree with a ring tuft, the search cost is the number of nodes N of the ordered tree (the number distinguished by the first hash value) and the number M of the ring tufts. The worst log (N) + M for (number distinguished by the second hash value).

  According to the present invention, it is possible to reduce the communication path length between management means for managing the communication matrix, etc., and further, it is possible to perform local repair and maintenance of the communication path at low cost. It becomes.

  In the present embodiment, an agent will be described as an example of the “communication matrix” of the present invention. Here, the agent is software that is placed on a platform machine such as WS or PC, receives a request from another agent or machine by a message or the like, and processes the request independently and in parallel.

  The agents in the present embodiment constitute a network having a multi-level structure (hereinafter referred to as “logical network”). The platform machines on which the agents are placed are connected to each other via an existing network such as TCP / IP (hereinafter referred to as “physical network”).

  FIG. 1 is a diagram showing a multi-layered structure of a logical network formed by agents in the present embodiment. Each agent has an address on the logical network (hereinafter referred to as “logical address”) by multi-level addressing from higher (wide area) to lower (local) along the hierarchical structure of the logical network, such as an address in the postal system. Is attached). For example, referring to FIG. 1, since agent a belongs to community B and higher community A, an address A.B.a is assigned to agent a. Similarly, the address A.B is assigned to the agent B. Here, the agent B behaves as if it is a single agent with respect to other communities or agents. An agent that behaves like Agent B is called a portal agent in the sense of a community window.

  The logical network is independent of the physical network, logical addresses are used for communication between agents, and the structure of the physical network is hidden.

  However, in practice, inter-agent communication on the logical network is performed on the physical network. In this embodiment, an agent communication zone (hereinafter referred to as “ACZ”) is provided on the platform machine, and this ACZ is an interface between the logical network and the physical network (from the logical address of the agent to the physical address of the ACZ). And conversion to the agent address in the ACZ), and performs route management on the physical network.

  Multiple ACZs may exist on a single platform machine. When an agent newly joins a logical network, a logical address (hereinafter also referred to as a name or ID) is assigned, and at the same time, an ACZ in which the agent operates is assigned. Therefore, precisely, each agent is operating on a specific ACZ. In consideration of this, in this embodiment, the platform on which the agent operates is called ACZ, and the physical network configured by ACZ is called an ACZ network.

  FIG. 2 is a diagram illustrating an example of an ACZ network to which the logical network illustrated in FIG. 1 is mapped. In FIG. 2, agent A is on ACZ01, agent AB is on ACZ03, agent ABC is on ACZ07, agent AD is on ACZ09, agent E is on ACZ12, and agent F is on ACZ13. FG runs on ACZ10, Agent FGH runs on ACZ15, and FGI runs on platform ACZ16. Agent ABa is on ACZ04, Agent ABCb is on ACZ05, Agents ABc and ADc are on ACZ06, Agents ABCd and ADd are on ACZ06, Agent Ae is on ACZ03, Agent Af is on ACZ02, Agent ADg On ACZ01, Agent Eh on ACZ12, Agent Ei on ACZ02, Agent Ej on ACZ04, Agent FGHk on ACZ14, Agent FGHl on ACZ15, Agent Gim on ACZ13, Agent Fn On ACZ11, Agent Fo is running on ACZ07.

  The structure of the ACZ network will be described with reference to FIG. The basic structure of the ACZ network is an ordered tree with a ring tuft (hereinafter referred to as “OTT”) in which a ring tuft is connected to a node of the order tree. In FIG. 2, the connection relationship (hereinafter referred to as “OTT link”) of this basic structure is indicated by a solid line, and the other connection relationship (hereinafter referred to as “bypass link”) is indicated by a broken line.

  FIG. 3 is a conceptual diagram of OTT, which is the basic structure of the ACZ network of FIG. This OTT has a physical address given to each ACZ (hereinafter referred to as “ACZ address”), and is configured using the value. The hash function employed at this time is a binary tuple (u, v), and the hash values of different ACZ addresses are different.

  The configuration of the ordered tree consists of the hash values of the ACZ addresses of the ACZ corresponding to each node (hereinafter, the nodes constituting the ordered tree are referred to as “node nodes”) and the left and right child node nodes, respectively ( When u, v) and (u1, v1) and (u2, v2), u1 <u <u2. Nodes having the same first hash value u and different second hash values v are connected to the annular tuft of each node node. Note that there is no ordering with respect to the second hash value v of the ACZ address of the corresponding ACZ with respect to the arrangement of the nodes connected to the annular tuft (hereinafter referred to as “annular tuft node”).

  FIG. 4 is a diagram in which the ACZ network of FIG. 2 is re-expressed based on the basic structure OTT. In FIG. 4, for example, the description “ACZ01 (8,3)” means that the hash value of ACZ01 is (8,3), and the same applies to other descriptions. The node nodes are ACZ01, ACZ02, ACZ04, ACZ07, ACZ09, ACZ10, ACZ12, and ACZ17. However, the annular chamber nodes ACZ14, ACZ15, and ACZ16 are connected to ACZ12, and the annular chamber node ACZ13 is connected to ACZ17.

  Some OTT nodes have additional ACZ bypass links. The bypass link is established between ACZs when peer-to-peer communication between agents (or between ACZs) occurs.

  FIG. 5 is a block diagram showing the ACZ 5 operating on the platform machine 1. Agents 3a to 3i operate on the ACZ5. The ACZ 5 includes a communication path search processing unit 7 that performs a communication path search process based on communication path establishment requests with other agents by the agents 3a to 3i, and a communication path search request from another ACZ. A duplicate search elimination unit 9 that determines whether the connection has been performed and a connection relationship management unit 11 that manages a connection relationship with another ACZ are provided.

  Information transfer on the ACZ network is performed by packet transfer between ACZs (bucket relay method). Each ACZ obtains the adjacent ACZ connected by the OTT link and the bypass link, and transfers information by tracing the adjacent ACZ from the transmission source ACZ to the transmission destination ACZ. In order to control this path, the connection relationship management unit 11 of each ACZ manages four types of tables.

  The first table is a local agent table 13 (hereinafter referred to as “LAT”). The LAT uses a combination of the name (ID) of an agent existing in its own ACZ and the physical address of the agent as one entry. It holds the physical address of the agent that exists in its own ACZ.

  The second table is a remote agent table 15 (Remote Agent Address Table, hereinafter referred to as “RAT”). In the RAT, a combination of a logical address (ID) of an agent existing in another ACZ and an ACZ address where the agent exists is defined as one entry. It holds the address of the ACZ where the agent that exists in another ACZ operates.

  The third table is an adjacent table 17 (ACZ Link Table, hereinafter referred to as “LT”). The LT takes a pair of the destination ACZ address and the reachable adjacent ACZ address as one entry, and holds the address of the adjacent ACZ on the route to the information transmission destination ACZ.

  The fourth table is an OTT adjacent link table 19 (Adjacent AT Table on OTT, hereinafter referred to as “AT”) that holds the links on the OTT of the parent, left child, right child, left adjacent to the annular chamber, and right adjacent to the annular chamber. ). The AT has a set of a parent ACZ address, a left child ACZ address, a right child ACZ address, a ring chamber left adjacent ACZ address, and a ring chamber right adjacent ACZ address as one entry. Since these links are adjacent links, they are also held in the LT, and in the implementation, the LT and AT are managed together in a single table, but here, in order to make the explanation easy to understand, the concept is used. Therefore, an AT is provided separately from LT.

  FIG. 6 is a diagram showing an example of the contents of the table held by ACZ01. (A) is an example of LAT, (b) is an example of RAT, (c) is an example of LT, and (d) is an example of AT. is there. Actually, each of these tables has a finite size, and it is impossible to completely hold all connection information of the network. Since the size of each table is set to a certain finite value, when a new ACZ is to be registered, if the size exceeds that size, a low-priority ACZ will be evicted from the table (priority level As the setting, for example, the number of times of communication in a certain period of time can be considered (Least Recently Used). For this reason, the link information once registered in the table does not always exist on the table. The present embodiment provides a robust network path management method that takes this into consideration.

  Next, communication between ACZs and path management between ACZs will be described. In this embodiment, Ack is exchanged between ACZs that perform communication, and the existence of the other party is mutually confirmed.

  Information transfer between ACZs is performed by packet transfer by following adjacent ACZ links of each ACZ on the communication path from the transmission source ACZ to the transmission destination ACZ. Note that it is assumed that an OTT link and a bypass link are registered in LT, and a link to an adjacent ACZ is provided by the OTT link and the bypass link. This communication path search process is performed by the communication path search processing unit 7 of FIG.

  When transferring information from one agent to another, if the destination agent is on the same ACZ (if it is registered in the LAT), it is passed to that agent. If the destination agent is on another ACZ (if it is not registered in the LAT but registered in the RAT), it is transferred to the destination ACZ and then passed to the destination agent after seeing the LAT on the destination ACZ. If it is not registered in the LAT or RAT, the ACZ searches for the destination agent and its route to the ACZ and creates a communication route between the two agents.

  First, a transfer method when a communication path has already been formed will be described. The structure of the transfer packet is [destination agent name, destination ACZ address, transmission information content]. The ACZ that has received this packet determines whether or not it is the destination ACZ, and if so, the destination agent is present on the own ACZ and knows its physical address by LAT. If it is not the destination ACZ, it looks at the LT, extracts the adjacent ACZ on the route to the destination ACZ, and transfers the packet to the adjacent ACZ.

  The message transfer using this LT will be described with reference to FIG. FIG. 7 is a diagram illustrating an example of information transfer from agent A.B.a on ACZ04 to agent E.h on ACZ12 and agent F.G.I.m on ACZ13 using LT.

  First, information transfer from agent A.B.a on ACZ04 to agent E.h on ACZ12 will be described. Agent A.B.a is on ACZ04 and sends a message on ACZ04. ACZ04 sees RAT and finds that Agent E.h is on ACZ12. Therefore, ACZ04 checks LT, finds that ACZ12 is adjacent, and forwards the message to ACZ12. ACZ12 looks at the LAT to confirm that Agent E.h is there and delivers the message to that physical address.

  Next, information transfer from agent A.B.a on ACZ04 to agent F.G.I.m on ACZ13 will be described. ACZ04 looks at the RAT and finds that Agent F.G.I.m is on ACZ13. Next, ACZ04 sees LT and knows that the adjacent ACZs on the path to ACZ13 are ACZ08, ACZ11, and ACZ12. Here, ACZ08 is selected first, and ACZ04 transfers a message to ACZ08. Upon receiving the transfer message, ACZ08 transfers the message to the adjacent ACZ06 from LT in the same manner as ACZ04. Similarly, ACZ06 forwards the message to ACZ13. ACZ13 knows that it is the destination of the message, confirms that Agent F.G.I.m is at LAT, and delivers the message to its physical address. The transfer path in this case is 3-hop ACZ04 → ACZ08 → ACZ06 → ACZ13.

  If for some reason ACZ08 has left OTT (Ack signal is not returned from ACZ08 to ACZ04), ACZ11 will be selected as the adjacent ACZ. In this case, the transfer path is ACZ04 → ACZ11 → ACZ06 → ACZ13. Become. Thus, by registering a plurality of adjacent ACZs and maintaining redundant transfer routes, the probability of performing a new route search can be lowered.

  If the destination agent is not registered in the source ACZ LAT or RAT, search for the agent, configure a route to the destination ACZ, and set the destination agent and its ACZ address in the source ACZ RAT. sign up.

  Agent search is performed by the following method. First, the communication path search processing unit 7 in FIG. 5 checks whether the search target agent or the ACZ in which the portal agent of the higher community to which the agent belongs is registered in the RAT. Here, since the agent name reflects the logical network address structure, the RAT search is performed with the longest prefix of the agent name prefix. The reason for taking the longest prefix match is that the matching agent is the lowest layer in the upper community in which the agent to be searched is a member, and the search space for the ACZ address of the agent to be searched is the smallest This is because it can be expected to be focused on.

  If the agent name exactly matches, you know the ACZ that the agent is running on.

  When the longest prefix part of the agent name matches, the address of the ACZ where the agent whose name is the prefix part is operating is known. This ACZ (hereinafter referred to as “search relay ACZ”) is requested to search for a search agent. Upon receiving the request, the search relay ACZ checks whether the search clue of the agent to be searched is registered in its own RAT. Similarly, the target agent ACZ is searched through the search relay ACZ.

  If the prefixes do not match, no search clue is registered in the RAT. Therefore, search for agents across the entire OTT. Each ACZ forwards an agent search message to all neighboring ACZs. The adjacent ACZ that has received the search message searches for the search target agent in the same manner.

  As an example, let us consider a case where an agent A.D.g currently operating in ACZ01 searches for another agent (see RAT of ACZ01 shown in FIG. 6B).

  When searching for agent A.D.d, A.D.d is registered in the RAT of ACZ01, and it can be seen that this agent is in ACZ06.

  When searching for agent A.B.c, it knows that agent A.B is operating in ACZ03 by the longest prefix match. Then, the search is requested with ACZ03 as the search relay ACZ.

  When searching for the agent F.G.H.k, the ACZ in which this agent and the portal agent of its higher community operate is not registered in the RAT. Therefore, a search message is transmitted to all adjacent ACZs (ACZ02, ACZ04, ACZ05, ACZ08, ACZ12) registered in LT and AT. (In this case, since ACZ01 is the root node of OTT, there is no parent.) Among those registered in LT, those not registered in AT (ACZ12) are bypass links ACZ. ACZ02, ACZ04, ACZ05, ACZ08, and ACZ12 each perform the same agent search process.

  When a search message is received from a child node, it is checked whether the agent to be searched is itself. If you are not yourself, send a search message to the left (or right) on the annular chamber and check whether the agent is in the annular chamber. If it is in the ring chamber, the search is terminated.

  If the message is received from the right (or left) child node when it is not in the ring tuft, the OTT subtree having the left (or right) child node as a root node is searched downward. At the same time, the search message is forwarded to the parent node of the OTT and the bypass link adjacent node. If it is received by a bypass link other than the parent and left and right children, an OTT subtree having the left and right child nodes as root nodes is searched.

  Since the search message is sent to all neighboring ACZs, the search spreads through the connection link. The search path length is at most 2 * log (n) + m with respect to the number n of nodes in the OTT and the ring circumference m of the annular tuft, but in practice the path length can be shortened by using a bypass link. When the bypass link is used in combination, there is a high possibility that the search target agent will be quickly found through a short search route through the bypass link, but the search route is multiplexed and a useless search occurs. In order to eliminate this waste, the search message is held in each node until the search is completed, and the duplicate search elimination unit 9 of each ACZ checks for multiple reception of the same search message.

  In addition, the search message has a search route stack, and an ACZ sequence that has been routed through the route search process is stacked. Further, the search message has an upper limit on the number of search hops to suppress the expanse of the search space. However, the upper limit of the number of hops is ignored in the upward and downward search of OTT in order to guarantee the completeness of the search.

  As a result of the route search, when the address of the ACZ where the agent to be searched exists is obtained, the destination ACZ registers the address of the source ACZ in the LT, and each ACZ on the route obtained by the search is adjacent to the LT. By registering the address of ACZ and the source ACZ registering the address of the returned destination ACZ in LT, the bidirectional path connecting the source ACZ and destination ACZ and both between the source ACZ and destination ACZ A directional bypass link is formed.

  The search using this search message will be specifically described. The structure of the search message is [search target agent name, search path stack, search source agent + ACZ, number of search hops]. The source ACZ sets an initial value in the search message and sends it to all adjacent ACZs. As the initial value of the search message, the agent name to be searched as the search target agent name is set to nil as the search path stack, the local agent name and the local ACZ address are set as the search source agent + ACZ, and the search hop count is set to 0. .

  Upon receiving the search message, the duplicate search elimination unit 9 of each ACZ checks whether the same search message has already been received, and if already received, discards this search message and ends.

  If the ACZ search message is received for the first time, the adjacent ACZ address of the transfer source is put on the search path stack of the search message. Then, looking at the LAT, it checks whether the search target agent is in its own ACZ. If it exists, the search is terminated. If it is not in the LAT, look at the RAT to see if there is a longest prefix match with the name of the search target agent, and if so, search for this as the search relay ACZ. If there is no search target agent in the RAT, the LT is seen and a search message is sent to the adjacent ACZ.

  When the search target agent ACZ is reached, the ACZ address is returned to the search source ACZ. The reply message follows the ACZ route sequence loaded in the search route stack in the reverse order and is transferred to the search source.

  The structure of the reply message is [search agent name + ACZ, search path stack, search source agent + ACZ]. In the transfer process of the reply message, each ACZ on the route registers the adjacent ACZ address in the LT and creates a bidirectional link to generate bidirectional route information between the search source ACZ and the search target ACZ.

  First, the search target ACZ registers the search source agent name and its ACZ address in the RAT, and registers the adjacent ACZ address and search source ACZ at the top of the search path stack in the LT.

  When each ACZ on the route receives a reply message, it registers the adjacent ACZ address and search target ACZ address at the top of the search route stack in LT, pops up the search route stack, and searches the source ACZ address and the search route stack. The adjacent ACZ address at the top of the is registered in the LT, a reply message is transmitted to the adjacent ACZ at the top of the search path stack, and the held search message is discarded.

  When the reply message reaches the search source ACZ, the search source ACZ registers the return source ACZ (search target ACZ) address and the adjacent ACZ address at the top of the search route stack in the LT, pops up the search route stack, and searches Register the target agent name and its ACZ address in the RAT, and discard the stored search message.

  Next, the search for the destination ACZ and the construction of the transfer path between ACZs when the path between the ACZs is cut off during system operation will be described. Neighboring ACZs that reach the destination include, for example, ACZ disconnection on the transfer path to the destination ACZ (platform machine power off or system down, agent or ACZ intentional network disconnection) Lost due to link disappearance (such as disappearance of adjacent ACZ address due to RAT or LT overflow). In this case, the transmission destination ACZ is searched using the ACZ that has lost the adjacent ACZ reaching the transmission destination as a new search source ACZ, and the transfer path is reconfigured. Each ACZ periodically checks whether there is an adjacent ACZ that reaches the destination registered in the LT, and if the adjacent ACZ is lost, searches for the destination ACZ and creates a new transfer route. Reconfigure.

  In the ACZ search, a hash value (u, v) of the address of the search destination ACZ is obtained, and a route to the transmission destination ACZ is searched using this hash value. In the search procedure, the search space on the logical network is narrowed down by the longest match of the agent ID prefix from the RAT in the agent search, whereas the OTT search space is narrowed down using the hash value in the ACZ search. It is almost the same as the agent search.

  First, the search message is transferred to the ACZ adjacent to the transmission source ACZ. The structure of the search message is [hash value (u, v) of search target ACZ, search path stack, search source ACZ, number of search hops]. The route search is performed using both OTT link and bypass link. In the search on the OTT link, if the binary set of the hash value of the search target ACZ is (u, v), the OTT node node equal to the hash value u is searched first, and then the circular cluster of the searched node nodes is searched. Is performed by searching for a node equal to the hash value v. In this search, it is known whether or not a search target exists, and if it exists, it always succeeds (the search completeness is guaranteed).

  In route search on OTT, the ratio of leaf nodes to all nodes is 50% from the structure of the ordered tree, and 75% when including the parent node of the leaf node (87.5% when including the parent) It is. When a route is formed by simply searching for an OTT route without using a bypass link, it goes from the node node close to the leaf node to the node close to the root node, and then goes down to the search target node close to the leaf node. The probability that a route is formed is high, and the length of most routes is log (n), which is a length that cannot be ignored. Therefore, as described below, the path length may be shortened by using the minimum value and the maximum value of the first hash value in the subtree for the path search on the OTT.

  Each ACZ holds its own first hash value and the minimum and maximum values of the first hash value in the OTT subtree having itself as a root node. Also, the first hash value of the adjacent ACZ registered in the LT and the minimum value and the maximum value of the first hash value of the OTT subtree having the node (ACZ) as the root node are stored in the LT. (In each LT entry, in addition to the address of the destination ACZ and the address of the reachable neighboring ACZ, the first hash value of the neighboring ACZ, and the set of the minimum and maximum values of the first hash value of the subtree Extend to hold.)

The operation of the search source or the node (ACZ) n0 that has received the search message will be described with reference to FIG. FIG. 8 is a diagram for explaining a search message transmission destination node. In FIG. 8, a symbol n (u, u _min , u _max ) attached to a node has a first hash value u for the node n, a minimum first hash value in OTT having this as a root node, u _min , It indicates that the maximum first hash value is u _max .

Let u be the first hash value of the search target ACZ, u0 be the first hash value of ACZ n0 that received the search source or search message, and u0 _{min be} the minimum value of the first hash value of the OTT subtree with itself as the root node The maximum value is u0 _max . If u = u0, find the node with v = v0 in the annular tuft of node node n0.

If u ≠ u0 and u0 _min ≦ u ≦ u0 _max , a downward search of the OTT subtree having itself as a root node is performed (see FIG. 8A). In the downward search, the minimum value of the first hash value of the OTT subtree of the bypass link destination is u _k-min and the maximum value is u _k-max , and u0 _min <u _k-min ≦ u _k <u ≦ u _{k The} search message is _sent to the node n _k having the smallest u _k-max −u _k-min among those satisfying _−max <u0 _max or u0 _min <u _k-min ≦ u <u _k ≦ u _k-max <u0 _max. send. Node n _k performs further downward search. If there is no such _nk , a search message is sent to the left child if u <u0, and to the right child if u0 <u. The left (right) child node that receives the message performs a downward search.

If u <u0 _min or u0 _max <u (ie, other than the above two cases), the subtree search is not performed, and if the current hop count is less than or equal to the hop count upper limit, the hop count is incremented by 1 and the current ACZ A search message is sent to all adjacent ACZs registered in the LT (see FIG. 8B), and the search is terminated if the current hop count exceeds the hop count upper limit.

  This OTT search is at most O (log n) with respect to the number of nodes n, but in actuality, it is expected to be much lower than O (log n).

  When the address of the destination ACZ is obtained by the route search, the bidirectional path that connects the source ACZ and the destination ACZ and the bidirectional bypass link between the source ACZ and the destination ACZ are sent to the LT. The address of the ACZ is registered, the address of the ACZ on both sides is registered in the LT of each ACZ on the route, and the source ACZ is formed by registering the address of the returned destination ACZ in the LT.

  The search operation by the OTT link and the bypass link will be specifically described.

  First, the source ACZ sets initial values of the search message [search target ACZ, search route stack, search source Agent + ACZ, number of search hops] (search ACZ address as search ACZ, nil as search route stack, The own ACZ address is set as the search source ACZ, and the upper limit value of the search hop count is set as the search hop count.) This is sent to all adjacent ACZs.

  The ACZ that has received the search message first checks whether the same search message has already been received. If it has already been received, the search message is discarded and the process ends. If the search message is received for the first time, the source adjacent ACZ address is loaded on the search message stack of the search message, the LT is seen, and the search message is transmitted to the adjacent ACZ. (The search on the OTT follows the algorithm described with reference to FIG. 8.)

  When the ACZ to be searched is reached, the ACZ address is returned to the search source ACZ. The reply message follows the ACZ path loaded in the search path stack in reverse order and forwards it to the search source. The structure of the reply message is [search target ACZ, search route, search source Agent + ACZ, (own first hash value, minimum first hash value of OTT subtree, maximum first hash value of OTT subtree)].

  In the return message transfer process, each ACZ on the path registers the adjacent ACZ address and its first hash value, the minimum first hash value of the OTT subtree, and the maximum first hash value of the OTT subtree in the LT, and both Bidirectional path information between the search source ACZ and the search target ACZ is generated by setting a directional link. The search target ACZ registers the source Agent name and ACZ address in the RAT, and the adjacent ACZ address at the top of the search path stack and its (first hash value, minimum first hash value of the OTT subtree, OTT part The maximum first hash value of the tree) and the search source ACZ are registered in the LT. When each ACZ on the route receives the reply message, it registers the adjacent ACZ address and return source ACZ address (that is, the search target ACZ) address at the top of the search route stack in LT, pops up the search route stack, and searches the search route stack. Register the adjacent ACZ address at the head of the first address and its (the first hash value of itself, the minimum first hash value of the OTT subtree, the maximum first hash value of the OTT subtree), and the search source ACZ address in the LT, Discard retained search messages. This is repeated until the reply message reaches the reply destination (search source) ACZ. When the reply message reaches the search source ACZ, the search source ACZ includes the adjacent ACZ address at the top of the search path stack, the address of the return source ACZ (search target ACZ), its first hash value, and the minimum first of the OTT subtree. The hash value and the maximum first hash value of the OTT subtree are registered in the LT, the search path stack is popped up, the search target agent name and its ACZ address are registered in the RAT, and the stored search message is discarded.

  Next, formation of an OTT link by the connection relation management unit 11 when an agent newly joins will be described. When an agent joins a community, it checks whether the ACZ on which the agent operates already exists in the OTT (this is the same method as the search process), and if it does not exist, the ACZ is set as a new node in the OTT. Connect. The addition of a new node and the maintenance of the maximum value and the minimum value of the first hash value in each node are performed as follows.

  Assuming that the first hash value of the new node to be added is u, search for the OTT node node n having the first hash value u as described above. If there is an OTT node node n whose first hash value is u, a new node is added to the ring tuft.

When there is no OTT node node n having the first hash value u, the node node that has received the new node addition message is set to n0, and the minimum value and the maximum value of the first hash value of the OTT subtree of this node are represented by u0. Let _min and u0 _max . At this time, if u0 _min ≤ u ≤ u0 _max , the new node is added to this OTT subtree, so the minimum value of the first hash value of the OTT subtree of the bypass link destination is set to u _{k- min} , where u _k-max is the maximum value, u0 _min <u _k-min ≦ u _k <u ≦ u _k-max <u0 _max or u0 _min <u _k-min ≦ u <u _k ≦ u _k-max < Among those satisfying u0 _max , a new node addition message is sent to n _k where u _k-max −u _k-min is minimum.

If u <u0 _min , the new node becomes the leftmost leaf node, so the first hash minimum value of n0 is replaced with u (u0 _{min is set} to u), and a new node addition message is sent to the left child. If the node that received this message is a leaf node, a new node is generated as a left child leaf node. If u0 _max <u, the new node becomes the rightmost leaf node, so the first hash maximum value of n0 is replaced with u (u0 _{max is set} to u), and a new node addition message is sent to the right child. . If n0 is a leaf node, a new node is generated as a right child leaf node.

FIG. 9 is a diagram illustrating an example of adding a new node. FIG. 9A shows a case where a leaf node is added as a left child, and uk _{-min of} ACZ0 is updated. FIG. 9B is a diagram showing an example of addition as a right child leaf node, and uk _{-max of} ACZ0 is updated.

  The addition of a new leaf node changes the maximum (or minimum) first hash value of the subtree held by its parent, and is updated. If it is necessary to change the maximum (or minimum) first hash value of the parent node by this change, the change is performed. Repeat this operation as necessary for ancestor nodes.

  Note that the OTT search and the generation of a new node are at most O (log n) with respect to the number n of nodes, but in reality, it is expected to be much lower than O (log n).

  Next, with reference to FIG. 10, when the adjacent link is disconnected due to the ACZ leaving the network, the re-establishment of the adjacent link by the connection relationship management unit 11 will be described. FIG. 10 is a diagram illustrating an example of an operation when the ACZ is detached.

  When the disconnection of the adjacent link occurs due to the voluntary disconnection from the ACZ network, the adjacent ACZ notifies the adjacent ACZ that the disconnecting ACZ will disconnect, so the adjacent ACZ that has received the disconnection notification will re-adjust the adjacent link. Do. However, withdrawal may occur due to an unexpected cause. Therefore, each ACZ periodically checks whether the connection link to the adjacent ACZ is valid, and if the adjacent link is broken, the adjacent link is re-established. This re-adjustment of the adjacent link will be described below. In some cases, an information transfer request to the transmission destination occurs before the link is re-established. In this case, the transmission destination search and the route reconstruction are performed as described above.

  When the leaving node is a circular atrio-node (see FIG. 10 (a)), the link between the left adjacent node m1 and the right adjacent node m2 of the leaving node n is established again. This is done by updating the addresses of the ATs of the annular left and right adjacent nodes m1 and m2.

  When the node node of the order tree in which the annular tuft node exists (see FIG. 10B), the node m on the right side of the tuft node of the leaving node node n is moved to the position of the node node n. The ATs of the node m, the circular left neighbor node m1, the left child node n1, the right child node n2, and the parent node n0 are updated to reestablish the adjacent link.

  When a node of an ordered tree having no annular tuft node leaves (see FIG. 10C), the rightmost descendant leaf node m1 of the left child node of the leaving node node n or the leftmost descendant leaf of the right child node. Move node m2 to position n. In particular, when there is no right child node, m1 (m2 when there is no left child node) is moved to the position n. Update the AT of parent node n0, left child node n1, and right child node n2 (m1 or m2) to reestablish the adjacent link

  When a node leaves, the first hash maximum value and minimum value of the OTT subtree of each node node that is an ancestor of the node node changes. At this time, if the maximum value changes to a smaller value or the minimum value changes to a larger value, the safety of the search is guaranteed without reflecting this change to each node. Do not change the maximum and minimum values. When the maximum value changes to a larger value or the minimum value changes to a smaller value, the maximum (minimum) first hash value of the parent node is updated.

  In this embodiment, each agent may perform file management and construct a distributed file system.

It is a figure which shows the multi-hierarchy structure of the logical network which the agent in a present Example comprises. It is a figure which shows an example of the ACZ network to which the logical network shown in FIG. 1 was mapped. FIG. 3 is a conceptual diagram of an ordered tree (OTT) with a ring tuft which is a basic structure of the ACZ network of FIG. It is the figure which re-expressed the ACZ network of FIG. 2 based on OTT which is the basic structure. 2 is a block diagram showing an ACZ 5 operating on the platform machine 1. FIG. It is a figure which shows an example of the content of the table which ACZ01 has, (a) is an example of LAT, (b) is an example of RAT, (c) is an example of LT, (d) is an example of AT. It is a figure which shows an example of the information transfer from Agent A.B.a on ACZ04 to Agent E.h on ACZ12 and Agent F.G.I.m on ACZ13 using LT. It is a figure explaining the transmission destination node of a search message. It is a figure which shows an example in the case of adding a new node. It is a figure which shows an example of the operation | movement at the time of detachment | leave of ACZ.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 Platform machine, 3a-3i agent, 5 agent community zone (ACZ), 7 Communication path search processing part, 9 Duplicate search exclusion part, 11 Connection relation management part

Claims (9)

In a network system in which a plurality of information processing devices are connected and communication between information processing devices is performed by a packet transfer method,
Each of the information processing apparatuses includes one or more management units that manage one or a plurality of communication bases, and each of the management units includes a communication path search processing unit that makes a search request for a communication path to another management unit. Have
Each management means is provided with identification information enabling identification with respect to other management means, and at least one of the management means having the same calculated value calculated by a predetermined calculation formula based on the identification information is saved. As a management means, the other management means as an annular tuft management means,
The node management means and the ring tuft management means of the same calculated value are connected in a ring,
The clause management means is connected to one or more of the clause management means of different calculated values;
The communication path search processing means of the node management means is requested to establish a communication path from the managed communication base to the communication base managed by the other management means, and the communication path to the other management means is established. If not, depending on the information held about the other management means,
Make a search request for the communication path to the annular chamber management means of the same calculated value,
A search request for a communication path is made to some or all of the clause management means of different calculated values to be connected, or
A search request for a communication path is made to the annular chamber management means of the same calculated value and the node management means of different calculated values to be connected.
Network system. The clause management means is one for each management means of the same calculated value, the clause management means constitutes an ordered tree with calculated values calculated by a predetermined calculation formula based on identification information,
The communication path search processing means of the clause management means, when holding the identification information of the other management means, the calculated value calculated by a predetermined calculation formula based on the identification information of the other management means When the calculated value is the same as the calculated value based on the identification information based on the identification information, a search request for the communication path is made to the annular chamber management means. A search request for a communication path is made to some or all of the value clause management means.
The network system according to claim 1.   When the other management means to be connected is disconnected from the connection relationship, the management means is in the annular connection relationship with the management means to be disconnected, and the other management means is in the annular connection relationship with the management means to be disconnected. A ring-shaped connection relationship is established, and when the management means to be detached is a node management means having a ring cluster management means, one of the ring cluster management means is on the order tree of the node management means to be detached If the connection relation is set to the position of, and the management relation is not provided, one of the management means located in descendants of the order tree of the management means to be removed is close to the calculated value of the management means to be removed. The network system according to claim 2, further comprising connection relation management means for generating a connection relation as a position on the order tree of the means. The management means has a connection relationship with other management means in addition to the connection relationship constituting the ordered tree with a ring tuft,
The communication path search processing means also makes a search request for a communication path to a management means having a connection relationship other than the connection relationship constituting the ordered tree with a ring tuft,
When the management unit receives a communication path search request, the management unit determines whether or not the communication path search request has been received redundantly, and performs a communication path search process only when there is no duplication. Having exclusion means,
The network system according to claim 2 or 3. There is a precious relationship between the communication mothers,
The communication path search processing means cannot reach the management means for managing the communication matrix that is the target of establishment of the communication path, and reaches the management means that manages the communication matrix that is in the possession of the communication matrix that is the target of establishment of the communication path. When possible, a search request for a communication path is made to a management means that manages a communication host having the most exclusive relationship with the other communication host.
The network system according to claim 1. An information processing apparatus in which communication with another information processing apparatus is performed by a packet transfer method,
Comprising one or more communication management means,
Each of the communication management means has communication path search processing means for making a search request for a communication path to other communication management means,
Each communication management means is provided with identification information enabling identification with respect to other communication management means, and the communication management means having the same calculated value calculated by a predetermined calculation formula based on the identification information are directly or the same. It is indirectly connected through the communication management means of calculated values, and at least one of the communication management means of the same calculated value is directly connected to the communication management means of different calculated values,
The communication management means responds to information held about the other communication management means when a communication path establishment request to the other communication management means is made and a communication path to the other communication management means is not established. ,
Make a search request for the communication path to the communication management means of the same calculated value,
Make a search request for a communication path to some or all of the communication management means of different calculated values, or
A communication path search request is made to the communication management means of the same calculated value and the communication management means of different calculated values.
Information processing device. An information transfer method in an information processing apparatus in which communication with another information processing apparatus is performed by a packet transfer method,
The information processing apparatus includes one or a plurality of communication management units, and each communication management unit includes a communication path search processing unit that requests a communication path search to another communication management unit,
Each communication management means is provided with identification information enabling identification with respect to other communication management means, and the communication management means having the same calculated value calculated by a predetermined calculation formula based on the identification information are directly or the same. It is indirectly connected through the communication management means of calculated values, and at least one of the communication management means of the same calculated value is directly connected to the communication management means of different calculated values,
When the communication management unit makes a communication path establishment request to the other communication management unit and the communication path to the other communication management unit is not established, according to information held about the other communication management unit , Request communication path search to communication management means of the same calculated value, request communication path search to some or all of communication management means of different calculated values, or the same calculated value An information transfer method including a step of making a communication path search request to a communication management unit and a communication management unit having different calculated values.   A program for causing a computer to realize the information transfer method according to claim 7.   A recording medium on which the program according to claim 8 is recorded.