JP2009134341A - Peer searching method and communication apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To obtain a peer searching method capable of reducing the fluctuation of processing time, and capable of rapidly performing the processing. <P>SOLUTION: The peer searching method allows the peers constituting a peer-to-peer network with a distributed hash table system adopted thereto to find a transmission destination or a transfer destination of a content request among transfer destination peer information. The method includes: a step (S107) of generating a search tree with a branch condition defined therein through the use of the hash values of the peers included in the transfer destination peer information, and determining whether or not the hash value of the key of the content satisfies the branch condition in order from the upper stage of the search tree; and a step (S108) of adopting the peer corresponding to the branch condition, which is determined to satisfy the branch condition and in the lowest stage, as a search result. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a peer search method and a communication device for searching for a communication device (peer) holding content to be searched in a peer-to-peer network in which content is distributed and managed.

  In recent years, various types of peer-to-peer networks have been constructed to perform specific services using IP (Internet Protocol) networks such as the Internet without changing the functions of existing network devices. The number and scale are increasing year by year. In a peer-to-peer network, the amount of content information increases in proportion to an increase in participating peers (communication devices). Therefore, in a large-scale peer-to-peer network with a large number of participating peers, it becomes difficult to centrally manage content with a server.

  Thus, in a large-scale peer-to-peer network, a content distribution management method is used in which content is distributed to each peer for management. One of the content distribution management methods is a distributed hash table technique (hereinafter referred to as DHT (Distributed Hash Table)). In DHT, a key for searching for content such as a title of the content is converted into a fixed-length bit string (hash) using a hash function, and a peer that should manage the hash is determined according to the hash value. In DHT, several implementation methods have been proposed depending on the difference in the representation method of the hash space created by the hash.

For example, Chord developed by Stoica et al. At the University of California, Berkeley (see Non-Patent Document 1 below) is a DHT implementation method that represents a one-dimensional hash space. In Chord, when the hash is represented by ^N bits, the hash space is regarded as a circular space by setting the hash value 0 next to the hash value 2 ^N −1.

  In the Chord network, each participating peer is given a unique hash. Each peer manages content that has a hash that is less than its own hash and that is greater than the hash of the neighboring peer that is smaller in the hash space. For example, in the case of a 3-bit (N = 3) one-dimensional hash space where the hash of the three peers is “0”, “1”, “3”, respectively, the peer with the hash “0” (hereinafter referred to as the hash) The peer having the m is called peer #m) manages only the content information in which the hash of the content key is “4”, “5”, “6”, “7”, “0”.

Further, as the number of participating peers increases, it becomes difficult to have a table for managing the transfer destinations in mesh form that can be connected to all of the participating peers. Therefore, in the Chord, each peer connects with only some of the peers. These peers are called Fingers, and a table that manages the Fingers is called a Finger table. Each peer is equal to or equal to a hash obtained by adding 2 ⁱ (i is an integer satisfying 0 ≦ i <N) to its own hash (hereinafter referred to as “Finger [i]”). Are managed in the Finger table, and transfer is performed only to peers included in the Finger table. In the above example of three peers, Peer # 0 manages only Peer 1 corresponding to i = 0 Finger and Peer # 3 corresponding to i = 1 Finger as transfer destinations. Can only transfer data. Similarly, peer # 1 manages peers # 3 and # 0 as transfer destinations and can transfer data only to these two peers. The peer 3 manages only the peer 0 as a transfer destination, and can transfer data only to the peer # 0.

  When each peer receives data, it first obtains a hash of the key of the content of the data, and compares whether or not the hash is managed by its own DHT. As a result of the comparison, if the hash is not managed by its own DHT, an appropriate peer is searched from the Finger table and obtained, and the received data is transferred.

  The specific processing contents of the peer search at this time will be described. Here, the notation [x, y) indicates an area of a hash space not less than x and less than y. Also, the hash of the content key is set to h. First, i = N−1 and y = the hash of its own peer. Next, x is a hash of Finger [i], and it is determined whether h is included in the area [x, y). If it is included, the peer of Finger [i] is determined as the transfer destination. If not, y is x and 1 is subtracted from i. Then, x is determined again as a hash of Finger [i], and it is determined again whether h is included in the area [x, y). If included, the peer of Finger [i] is set as the transfer destination. If not, y is x and 1 is subtracted from i. Then, the transfer destination of the peer is determined, or the process is repeated by decreasing the value of i by 1 until i = 0. If h is not included in the area [x, y) even if i = 0, the transfer destination peer is the peer of Finger [0].

Since 2 ^N −1 and 0 are connected in the hash space, x may be larger than y. In this case, instead of determining whether h is included in the region [x, y), it is determined whether h is included in [x, 2 ^N −1] or [0, y).

  For example, when the peer # 1 wants to find a peer that manages the content whose hash of the content key is “4”, Finger [2] = Peer 3 is determined as the transfer destination by the above search process, and the peer # 1 transfers data to peer # 3. Peer # 3 performs the above-described search process, and even when i = 0, h is not included in the area of [x, y). Therefore, Peer # 0 that is Finger [0] is determined as the transfer destination, Transfer data to peer # 0. Finally, the peer 0 that has received this data recognizes that the content of the hash (key hash) “4” is in its own DHT, so that it is the content that it manages, searches for the corresponding content, and peer # Send to 1

  A peer-to-peer network may also be used when sending content to a requesting peer. Also in this case, the requesting peer is found by the above-described search process. In the above example, peer # 0 performs the above-described search process to send content to peer # 1 whose hash is “1”, and sends content to peer # 1 that is Finger [0] of peer # 0. To do. Peer # 1 obtains this content because the peer that requested the received content data matches the peer.

  In this way, by using the Finger table, in a peer-to-peer network where the number of participating peers is K, it is possible to obtain arbitrary content by the number of O (logK) transfers. Also, when there are a large number of peers participating in the peer-to-peer network and peer hashes and content hashes are evenly distributed, it is determined whether one peer is included in the management area of Finger [i]. The expected value of the number of times (region comparison) can be obtained by Σ {probability of content key hash in Finger [i] × number of comparisons reaching Finger [i]}. The larger i is, the smaller the number of comparisons and the larger the area occupied in the hash space, the higher the probability of existence of the hash, and the maximum expected value is 2. In an actual peer-to-peer network, there may be a case where Finger [i] and Finger [i + 1] overlap (Finger [i] and Finger [i + 1] are the same peer). The expected value is smaller than twice.

  Also, Chord employs Secure Hash Algorithm 1 (hereinafter referred to as SHA-1) as a hash function. SHA-1 is a standard hash function adopted by the United States Standards Technology Agency for the US government, and is used for authentication of electronic documents, detection of falsification, and the like. Since SHA-1 generates a 160-bit hash, the Chord Finger theoretically manages a maximum of 160 peers. In SHA-1, region comparison is performed between positive integers of 160 bits. At the present time, there are no special processors that can perform 160-bit integer comparison with one instruction. For this reason, a region comparison is executed using a plurality of instructions, and as the number of comparisons increases, the processing load is such that the processing speed is affected.

  On the other hand, even on a peer-to-peer network, there is an increasing demand for using a service that requires a high transmission rate, as represented by video distribution. As described above, since it is necessary to transfer data even in a peer other than the peer requesting the content or the peer managing the content, the transmission rate greatly depends on the transfer performance of each peer. In order to increase the transmission rate of the service, it is necessary to efficiently execute the data transfer process at each peer.

I. Stoica, R.A. Morris, D.M. Karger, M.A. F. Kaashoek and H. Balakrishnan, “Chord: A scalable peer-to-peer lookup service for Internet applications”, In Proceedings of ACM SIGCOMM 2001, San Diego, CA.

  However, according to the above-described conventional technique, the processing time differs in the conventional technique because the number of area comparisons in the peer peer-to-peer network that uses a one-dimensional hash space such as Chord differs depending on the hash value. . For this reason, in the case of an application that receives data at regular intervals, such as a voice service, there is a problem that processing time fluctuates and data quality deteriorates. For example, even if audio data is transmitted at a fixed interval at the source peer, if the peer that relays the data also transfers content other than the audio data, the search processing is performed for each content. As a result, the interval at which the audio data is output fluctuates. In that case, at the peer requesting the audio data (reception peer), depending on the decoding method, the audio reproduced for a time corresponding to the audio data delayed in reception may be silenced. As the number of peers to be relayed increases, the data arrival interval at the receiving peer greatly fluctuates and the voice quality deteriorates.

  As a measure for preventing such fluctuations in the arrival interval of data, a method of ensuring, as a search processing time, each peer always performs 160 area comparisons, which is the maximum number of area comparisons in SHA-1. There is. However, this method increases the delay even if there is no fluctuation. When the delay becomes large, the voice is heard with a delay, so that there is a problem that the other party's reaction becomes slow and uncomfortable for the caller.

  In order to reduce the delay, there is a method of reducing the processing time by parallelizing the processing. However, as described above, since the expected value of the number of comparisons is two, SHA-1 performs 160 region comparisons. The processing time is parallelized to 80 and the processing time is equal to the expected value. However, 80 parallel processing is not realistic. Therefore, even if a certain amount of parallel processing is performed, there is a problem that if measures are taken to prevent fluctuations, the delay becomes larger than when measures are not taken.

  Also consider a service that is not affected by the data arrival interval, such as a file transfer service. Many file transfer services use the TCP / IP protocol in IP networks. TCP / IP has a flow control function for changing a transmission rate according to transmission characteristics between peers. This function acts to lower the transmission rate of the source peer as the data arrival delay increases. For this reason, there is a problem in that the data transfer rate using TCP / IP becomes low if measures are taken to prevent such fluctuations as described above.

  The present invention has been made in view of the above, and in a peer-to-peer network in which content is distributedly managed, fluctuations in processing time of search processing of the peer managing the content are reduced, and the peer managing the content is managed. It is an object of the present invention to provide a peer search method and a communication device that can perform search processing at high speed.

  In order to solve the above-described problems and achieve the object, the present invention is based on a hash value assigned to each of the peers constituting a peer-to-peer network that adopts a distributed hash table method to distribute and manage content. When a data transfer destination peer determined in this way is managed as transfer destination peer information, a peer search for searching for a content request transmission destination or a transfer destination of the content request from peers included in the transfer destination peer information A search tree generation step of generating a search tree in which a branch condition is defined using a transfer destination peer hash value that is a hash value of a peer included in the transfer destination peer information, and corresponding to the content request A determination step for determining whether the hash value of the content key satisfies the branch condition in order from the top of the search tree. The transfer destination peer hash value corresponding to the branch condition located at the bottom of the search tree among the branch conditions determined to satisfy the branch condition is obtained, and the peer corresponding to the transfer destination peer hash value is searched. And a resulting peer determination step.

  According to this invention, it is determined whether or not the hash of the content key satisfies the branch condition using the search tree in which the branch condition is determined using the hash value of the transfer destination peer, and corresponds to the hash satisfying the branch condition. The search result is the lowest peer in the search tree among the peers that perform the search, so that fluctuations in the processing time of the search process of the peer managing the content can be reduced and the search process of the peer can be performed at high speed There is an effect that can be.

  Embodiments of a peer search method and a communication device according to the present invention will be described below in detail with reference to the drawings. Note that the present invention is not limited to the embodiments.

Embodiment 1 FIG.
FIG. 1 is a diagram illustrating an arrangement example of peers (communication devices) in the one-dimensional hash space of the network system according to the first embodiment of the present invention. A one-dimensional hash space 1 in FIG. 1 represents a 6-bit hash space. The network system according to the present embodiment is a peer-to-peer network that assigns a 6-bit hash to each peer. In this embodiment, it is assumed that content is managed in a distributed manner, and that each peer manages content whose content key hash is within the range of its own DHT.

  The peer 2 in FIG. 1 is a peer that searches for content, and is assigned a hash “30”. The Finger peer 3 is a peer included in a Finger table managed by the peer 2. In FIG. 1, only the peers 2 and 2 managed by the Finger table are shown, but the entire network system according to the present embodiment includes other peers in the one-dimensional hash space 1. And

  FIG. 2A is a diagram illustrating an example of a DHT managed by the peer 2. The peer 2 manages the content in which the hash of the content key is included in the DHT range. Therefore, in this case, the peer 2 manages content in which the hash of the content key is in the range of “26 to 30”.

In this embodiment, a binary search tree is used to search for peers that manage desired content. FIG. 2-2 is a diagram illustrating an example of the binary search tree of peer 2. In this embodiment, a binary search tree is generated based on the hash of each peer included in the Finger table. In this embodiment, the Chord method is adopted as the DHT implementation method, and is equal to a hash obtained by adding 2 ⁱ (i is an integer of 0 ≦ i <N: N is the number of bits of the hash) to the own hash, Only the nearest peer with a larger hash is managed by the Finger table. The upper diagram of FIG. 2-2 shows the Finger table of Peer 2 and shows, for each ⁱ , a hash obtained by adding 2 ⁱ to the hash of its own (Peer 2) as a hash of Finger [i]. Hereinafter, i in the Finger table is referred to as a Finger identification number. In FIG. 2A, each node of the binary search tree shown in the figure below is represented by a hash value. Actually, however, the binary search tree is managed by the binary search tree for both the Finger identification number and the corresponding hash. Suppose that it is done. The binary search tree generation method may be generated by any method as long as it is a normal search tree generation method.

  Next, a process (peer search process) for searching for a peer managing desired content according to the present embodiment will be described. FIG. 3 is a flowchart illustrating an example of a peer search processing procedure according to the present embodiment.

  In the conventional peer search processing, for each Finger, it is determined whether or not the hash of the content key is included in the content range managed by each Finger. In the present embodiment, in order from the top of the binary search tree, it is determined whether or not the hash of the content key is a value less than the hash of each node constituting the binary search tree (condition is satisfied). Then, the Finger corresponding to the last node for which the condition is satisfied is obtained as a peer to which the content is transmitted.

  Peer search processing according to the present embodiment will be described with reference to FIGS. First, the peer 2 obtains a hash of the key of the desired content and sets it as h (step S101). Next, the peer 2 determines whether or not h is in its own DHT (step S102). If it is determined that h is in its own DHT (Yes in step S102), it is determined that the content is managed by itself (step S103), and the peer search process is terminated.

  When it is determined that h is not in its own DHT (No in step S102), the variable dst is initialized (for example, NULL is substituted) (step S104). dst is a numerical value indicating the hash of the Finger obtained as the transmission destination of the content request. Next, the Finger identification number (i in the Finger table of FIG. 2-1) corresponding to the root node of the binary search tree shown in FIG. 2-2 is substituted for p (step S105). p is a variable indicating a Finger identification number of a Finger that is a determination target of condition establishment in Step S107 described later. Next, the peer 2 substitutes Finger [p] for the variable x (step S106).

  Next, the peer 2 determines whether or not h is smaller than x (step S107). When it is determined that h is not smaller than x (step S107 No: condition is satisfied), x is substituted into dst ( Step S108). Then, it is determined whether p is the lowest node in the binary search tree (step S109). If it is determined that p is not the lowest node (No in step S109), The finger identifier of the node is substituted for p (step S110), the process returns to step S106, and the processes after step S106 are repeated.

  If it is determined in step S107 that h is smaller than x (step S107 Yes: condition is not satisfied), it is further determined whether p is the lowest node in the binary search tree (step S111). ). If it is determined in step S111 that the node is not the lowest node (No in step S111), the finger identifier of the lower left node of the binary search tree is substituted for p (step S112), and the process returns to step S106, and after step S106. Repeat the process.

  If it is determined in step S109 that the node is the lowest node (step S109 Yes), and if it is determined that the node is the lowest node in step S111 (step S111 Yes), is dst still initialized? If it is determined whether it has not been initialized (step S113 No), the search process is terminated. If it is determined in step S113 that dst remains initialized (Yes in step S113), the maximum value (maximum Finger) of Finger [i] on the Finger table is substituted for dst (step S114), and the process is performed. finish.

  When the peer 2 determines that the content is managed in step S103 after the above peer search process (Yes in step S103), the peer 2 selects the content based on the information of the content managed by the peer 2. get. In other cases, the peer 2 transmits a content acquisition request (including the content key) to the Finger corresponding to dst at the end of the search process (the value of dst is assigned as a hash).

  Taking the case of h = 31 as an example, the processing of FIG. 3 will be specifically described. Since “31” is not in the DHT of the peer 2, Steps S104 to S107 are performed. In step S107, "31" is smaller than the root node hash "36", so the condition is not satisfied. At this point, p is not the lowest level, so steps S111 and S112 are performed, and p is set as the lower left node, and the process returns to step S106. In the second step S107, since h is not smaller than the hash “31” of the second node, the condition is satisfied, and the process proceeds to step S108. Since p is not the lowest stage at this point, Steps S109 and S110 are performed, and the process returns to Step S106 with p as the lower right node. In the third step S107, since h is smaller than the hash “32” of the third node, the condition is not satisfied, and the process proceeds to step S111. At this point, p is at the lowest level, and the process further proceeds to step S113. In the determination in step S113, it is determined that dst has not been initialized (No in step S113), and thus the process ends. At the end of the process, the hash “31” of the node that finally satisfies the condition is assigned to dst, and the peer corresponding to the hash “31” becomes the transmission destination of the content request.

  Next, the process of FIG. 3 will be specifically described in the same manner, taking the case of h = 2 as an example. Since “2” is not in the DHT of peer 2, steps S104 to S107 are performed. In step S107, “2” is smaller than the root node hash “36”, so the condition is not satisfied. At this point, p is not the lowest level, so steps S111 and S112 are performed, and p is set as the lower left node, and the process returns to step S106. In step S107 for the second time, h is smaller than the hash “31” of the second node, so the condition is not satisfied again. Since p is not the lowest stage at this time, Steps S111 and S112 are performed, and the process returns to Step S106 with p as the lower left node. In the third step S107, since h is smaller than the hash “5” of the third node, the condition is not satisfied and the process proceeds to step S111. At this point, p is the lowest node (step S111 Yes). Proceed to step S113. In this example, dst has never been updated and remains at the initial value. Therefore, in step S113, it is determined that dst remains the initial value (step S113 Yes), and in step S114, the maximum Finger is substituted for dst.

  As h = 2, h smaller than the smallest hash in the Finger table exists on the one-dimensional hash space 1 between the largest hash and the smallest hash in the Finger table across the hash 0. To do. For this reason, in the present embodiment, when the condition has never been satisfied in the determination in step S107 (dst remains the initial value), the destination of the content request is the maximum hash in the Finger table. A peer corresponding to 48 ″.

  FIG. 4A is a diagram illustrating the concept of the peer search process of the present embodiment when h = 31, and FIG. 4B is the concept of the peer search process of the present embodiment when h = 2. FIG. In FIGS. 4A and 4B, a node where the condition is satisfied is indicated by ◯, and a node where the condition is not satisfied is indicated by X.

  In the above description, the example in which the peer 2 selects the transmission destination of the content request has been described. However, the content request is not always transmitted directly to the peer that manages the content. Therefore, the peer that has received the content request performs the peer search process described above in the same manner, and if it is determined that the content request is not a request for content managed by the peer, the peer search process determines the transmission destination. Forward the content request to the peer. By repeating such an operation, the content request is finally transmitted to the peer that manages the content requested from the peer 2.

  In the present embodiment, the processing for acquiring content has been described. However, when a peer managing content transmits the content to the request source (peer 2 in the above example), When searching for a requesting peer, processing similar to peer search processing may be performed to search for a peer to which content should be transmitted.

  In this embodiment, an example in which peer search processing is performed using a binary search tree has been described. However, the present invention is not limited to this, and any search tree may be used as long as the method uses a search tree. It may be used. For example, in the case of a quadrant search tree, the hash for determining whether the condition is satisfied is three numerical values, and these are z (1), z (2), and z (3). At this time, for example, when h is smaller than z (1), the condition is not satisfied. When h is equal to or greater than z (1) and less than z (2), the condition is satisfied with respect to z (1). When z (2) or more and less than z (3), the condition is satisfied with respect to z (2), and when h is equal to or greater than z (3), the condition is satisfied with respect to z (3). You should make it go.

  As described above, in this embodiment, a search tree for searching a finger is used to define a hash range for establishing a condition for each node of the search tree, and the search tree in the nodes that satisfy the condition is determined. The peer corresponding to the lowest node is selected as the content request destination. For this reason, processing can be performed at a higher speed than the conventional peer search processing. In addition, a difference in processing time (fluctuation) due to a difference in the value of h can be reduced.

  For example, if the Finner binary tree is balanced (not a necessary condition), the maximum number of hash comparisons of the present method in the Nbit one-dimensional hash space can be suppressed to Round (log2 (N + 1)). (Round is a function of rounding up). Compared with the 160-bit one-dimensional hash space of Chord, the maximum number of comparisons in the prior art was 160 times (actually doubled because the comparison is made at both ends of the area). Then, the maximum number of comparisons can be reduced to 9.

Embodiment 2. FIG.
FIG. 5 is a flowchart showing an example of a peer search processing procedure according to the second embodiment of the present invention. The configuration of the network system of the present embodiment is the same as that of the first embodiment. Constituent elements similar to those in Embodiment 1 are denoted by the same reference numerals, and description thereof is omitted. The Finger table and binary search tree of peer 2 in the present embodiment are the same as those in the first embodiment.

  As shown in FIG. 5, step S101 and step S102 are the same as in the first embodiment. If it is determined in step S102 that h is included in the own DHT (Yes in step S102), the process proceeds to step S103 as in the first embodiment, and the process ends. If it is determined in step S102 that h is not included in its own DHT (No in step S102), the maximum Finger is substituted for dst (step S201). And step S105-step S112 similar to Embodiment 1 are implemented. However, if it is determined in step S111 that p is the lowest level (step S111 Yes), and if it is determined in step S109 that p is the lowest level (step S109 Yes), the steps of the first embodiment are performed. S113 and S114 are not performed, and the process is terminated. Other processes in this embodiment are the same as those in the first embodiment.

  Taking the case of h = 31 as an example, the process of FIG. 5 will be specifically described. Since “31” is not in the DHT of peer 2, the maximum Finger is substituted for dst (step S 201). And step S105-step S107 are implemented. In step S107, "31" is smaller than the root node hash "36", so the condition is not satisfied. At this point, p is not the lowest level, so steps S111 and S112 are performed, and p is set as the lower left node, and the process returns to step S106. In the second step S107, since h is not smaller than the hash “31” of the second node, the condition is satisfied, and the process proceeds to step S108. Since p is not the lowest stage at this point, Steps S109 and S110 are performed, and the process returns to Step S106 with p as the lower right node. In the third step S107, since h is smaller than the hash “32” of the third node, the condition is not satisfied, and the process proceeds to step S111. At this point, p is at the lowest level, so the process is terminated. At the end of the process, the hash “31” of the node that finally satisfies the condition is assigned to dst, and the peer corresponding to the hash “31” becomes the transmission destination of the content request.

  Next, the case of h = 2 is taken as an example, and the processing of FIG. Since “2” is not in the DHT of the peer 2, the maximum Finger is substituted for dst (step S 201). And step S105-step S107 are implemented. In step S107, “2” is smaller than the root node hash “36”, so the condition is not satisfied. At this point, p is not the lowest level, so steps S111 and S112 are performed, and p is set as the lower left node, and the process returns to step S106. In step S107 for the second time, h is smaller than the hash “31” of the second node, so the condition is not satisfied again. Since p is not the lowest stage at this time, Steps S111 and S112 are performed, and the process returns to Step S106 with p as the lower left node. In the third step S107, since h is smaller than the hash “5” of the third node, the condition is not satisfied, and the process proceeds to step S111. At this point, p is the lowest node (step S111 Yes), so The process is terminated as it is. At the end of processing, since the maximum Finger is still substituted for dst, the peer corresponding to “48” of the maximum Finger becomes the transmission destination of the content request.

  FIG. 6A is a diagram illustrating the concept of peer search processing according to the present embodiment when h = 31, and FIG. 6-2 illustrates the concept of peer search processing according to the present embodiment when h = 2. FIG. In FIGS. 6A and 6B, a node where the condition is satisfied is indicated by ◯, and a node where the condition is not satisfied is indicated by X.

  Thus, in this embodiment, a search tree for searching a finger is used to define a hash range for establishing a condition for each node of the search tree, and as an initial value of a hash value of a destination peer After setting the maximum Finger, the peer corresponding to the lowermost node among the nodes that satisfy the condition is selected as the content request transmission destination. For this reason, the same effect as in the first embodiment can be obtained, and the processing can be reduced as compared with the first embodiment. When the processing of the present embodiment is realized by hardware, the processing circuit of steps S113 and S114, which is necessary for the first embodiment, is unnecessary, and it is only necessary to give an initial value, so that the entire circuit configuration Is simplified, and hardware implementation becomes easy.

Embodiment 3 FIG.
FIG. 7 is a diagram illustrating an arrangement example of peers (communication devices) in the one-dimensional hash space of the network system according to the third embodiment of the present invention. A one-dimensional hash space 1a in FIG. 7 represents a 15-bit hash space. The network system of the present embodiment is a peer-to-peer network that assigns a 15-bit hash to each peer. In this embodiment, it is assumed that content is managed in a distributed manner, and that each peer manages content whose content key hash is within the range of its own DHT. In the hash space shown in FIG. 7, the following description will be made assuming that about 30 to 40 peers are participating in the network.

  In the present embodiment, it is assumed that the hash “6000h” (h indicates a hexadecimal number) is assigned to the peer 2 that requested the content. FIG. 8 is a diagram illustrating an example of the Finger table of the peer 2 according to the present embodiment. The hash of peers participating in the network is rarely concentrated in a specific range and is often distributed. Therefore, when the hash space is divided into ranges of every “1000h”, 3 to 6 units are included in each range. Often exist. Therefore, when i of Finger [i] is small, Finger [i] corresponding to a plurality of i often becomes the same.

For example, in the example of FIG. 8, ⁰ 2 to a hash value of the peer 2, 2 ^1, ..., absent peer with a hash of between 2 ⁸ corresponds to a range obtained by adding each "6001h" ~ "6100h", The peer having the hash value closest to the peer 2 is the peer whose hash is “61F0h”, and therefore, Finger [0] to Finger [8] are “61F0h”. Similarly, Finger [9] and Finger [10] are the same.

  On the other hand, in the 15-bit hash space, the Finner binary search tree has a maximum depth of 4 levels. In the example of FIG. 8, i with the same Finger [i] increases, so the actual number of peers of Finger [i] is less than 15. In this embodiment, even in such a case, the peer that becomes the maximum hash is duplicated as a dummy so that the number of nodes (peers) constituting the binary search tree is 15. That is, in this embodiment, the binary search tree is a balanced tree. An example of the binary search tree created in this way is shown in FIGS. In FIGS. 9A and 9B, a node that satisfies the condition is indicated by a circle, and a node that does not satisfy the condition is indicated by a cross. FIG. 9-1 is an example when h = 7F00h, and FIG. 9-2 is an example when h = 21E3h.

  The peer search process of the present embodiment is the same as that of the first embodiment described with reference to FIG. The operations other than the content search process are the same as those in the first embodiment except for the above-described binary search tree generation method. Taking the case of h = 7F00h as an example, the peer search processing of this embodiment will be specifically described. It is assumed that “7F00h” is not included in the DHT of peer 2. First, step S101 and step S102 of FIG. 3 are performed. In step S102, h is determined not to be included in its own DHT (No in step S102), and steps S104 to S107 are performed. In step S107, since h is equal to or greater than the hash “7321h” of the root node, the condition is satisfied, and the process proceeds to step S108. Since p is not the lowest stage at this point, Steps S109 and S110 are performed, and the process returns to Step S106 with p as the lower right node. Since h is also compared with “7321h” in step S107 for the second time, steps S108 to S110 are performed as in the first time, and the process returns to step S106. Thereafter, the conditions are also satisfied in the third and fourth steps S107, and after the fourth step S107, steps S108 and S109 are performed, and then the process proceeds to step S113 and the process is terminated. At the end of the process, the peer corresponding to the hash “7321h” that finally satisfies the condition becomes the transmission destination of the content request.

  Further, the case of h = 21E3h is as follows. It is assumed that “21E3h” is not included in the DHT of peer 2. First, step S101 and step S102 of FIG. 3 are performed. In step S102, h is determined not to be included in its own DHT (No in step S102), and steps S104 to S107 are performed. In step S107, since h is less than the root node hash “7321h”, the condition is not satisfied, and the process proceeds to step S111. And after implementation of step S111 and step S112, it returns to step S106. In the second step S107, since h is less than “66BAh” of the second-stage node, the condition is not satisfied, and the process proceeds to step S111. And after implementation of step S111 and step S112, it returns to step S106. In step S107 for the third time, since h is equal to or greater than “21E3h” of the third-stage node, the condition is satisfied and the process proceeds to step S108. Since p is not the lowest stage at this point, Steps S109 and S110 are performed, and the process returns to Step S106 with p as the lower right node. In the fourth step S107, since h is less than “61F0h” of the fourth-stage node, the condition is not satisfied, and the process ends after steps S111 and S113. Then, the peer corresponding to the hash “21E3h” that is finally established at the end of the process becomes the transmission destination of the content request.

  In the present embodiment, the time required for the content search process is constant without depending on the content and the hash of the peer, and not depending on the number of fingers. FIG. 10A is a diagram illustrating an example of parallel processing timing when the peer search processing according to the present embodiment is used. FIG. 10B is a diagram of an example of parallel processing timing when the conventional peer search process is used. As illustrated in FIG. 10A, it is assumed that each peer outputs the received peer-to-peer network data by performing parallel processing with three threads (thread # 1 to thread # 3). Further, data of content A is data A, and data of content B, which is content different from content A, is data B. At this time, assuming that the processing time of the peer search process is constant, as shown in FIG. 10A, the transmission interval of data A is substantially equal to the reception interval, and there is an effect that the arrival interval does not fluctuate. .

  On the other hand, even in the example using the content search process shown in FIG. 10-2, as in FIG. 10-1, each peer outputs the received peer-to-peer network data in parallel using three threads. To do. At this time, in the case of the example of FIG. 10A, the peer search processing time is not constant because it depends on the hash. Here, it is assumed that the peer search processing time for data A is shorter than the peer search processing time for data B. In this case, while the thread # 2 is in the process of transmitting the data B (before the transmission process is completed), the thread # 3 completes the peer search process. Therefore, when the network ports used by the thread # 2 and the thread # 3 overlap, the thread # 3 waits for the completion of the data B transmission process of the thread # 2, and starts the data A transmission process. Since this waiting time occurs, the transmission interval of the data A becomes longer than the reception interval. Therefore, there arises a problem that fluctuation occurs in the reception interval. On the other hand, as described above, the reception interval does not fluctuate in this embodiment.

  By the way, when considering the data delay time, the conventional method may have a shorter processing time and a shorter data delay time. As described above, in the Nbit hash space, the expected value of the number of times of performing the area comparison of the conventional peer search process is 2 × 2 (2 comparisons in 1 area), whereas in this search method, This is because the determination of the satisfaction of the condition of Round (log (N + 1)) times is always made, so in the Chord method with N = 160, the number of determinations is 9, which is larger than the expected value of the conventional number of area comparisons.

  However, in the peer search process of the present embodiment, the parallelization effect can be obtained linearly. That is, for example, when the parallelization is performed as shown in FIG. 10A, the processing time can be reduced to one third. In the case of making a determination nine times at the maximum, parallel processing corresponds to three times of processing, so that it is possible to improve processing speed as compared with the conventional method while preventing occurrence of fluctuation. In addition, since the processing time is constant, the hardware circuit can eliminate the need for a circuit such as a waiting buffer, and thus parallelization can be realized more easily than the conventional method.

  FIG. 11 is a diagram illustrating a functional configuration example of a search processing device that implements the peer search processing of the present embodiment by hardware. FIG. 11 shows only components that perform processing corresponding to steps S106 and S107 (hereinafter referred to as comparison processing) in FIG. As shown in FIG. 11, the search processing device includes a comparison circuit (determination means) 4, a memory 5, and a memory bus 6.

  The Finger table of each peer is stored in the memory 5. The comparison circuit 4 reads out one finger information (Finger [i]) by transmitting a read command via the memory bus 6. It is assumed that the information on Finger is stored in association with the position of Finger [i] in the binary search tree. When the comparison circuit 4 performs the comparison process of FIG. 3, the address corresponding to Finger [p] is designated and a read command is transmitted to the memory 5 via the memory bus 6, and a read response is transmitted from the memory 5 via the memory bus 6. And obtain the value of Finger [p] as x. Then, h and x are compared. Here, the processing of the comparison circuit 4, the memory 5, and the memory bus 6 can be pipelined. When pipelined, in this configuration example, it is possible to perform processing at a speed three times that when not pipelined. FIG. 12 is a diagram illustrating an example of processing when pipeline processing is performed.

  In the present embodiment, since the processing time is fixed, the order in which the results of the comparison processing are output is not changed even if the processing is pipelined. Therefore, it is not necessary to provide an order guarantee circuit in the subsequent stage, and it is not necessary to send a signal notifying that the search is completed and an input may be sent from the comparison circuit 4 to the previous stage. Further, a waiting circuit in the previous stage is not necessary. This facilitates the realization of hardware capable of high-speed processing by pipelining.

  In the present embodiment, the case where peer search processing similar to that in Embodiment 1 is performed with the binary search tree as an equilibrium tree has been described. However, the binary search tree is configured as an equilibrium tree. You may make it perform the peer search process of the form 2. Also in this case, since the comparison processing is the same as that in the first embodiment, the processing time is constant, and high-speed processing by pipelining can be easily realized.

  As described above, in this embodiment, a binary search tree is generated by adding a node corresponding to the maximum hash for the insufficient nodes so that the number of nodes constituting the binary search tree becomes the maximum number of nodes. Similar to the first embodiment, peer search processing is performed according to a binary search tree. For this reason, the processing time of peer search processing is constant, and pipelining is facilitated, so that the processing speed can be further increased as compared with the first embodiment.

  As described above, the communication device and the peer search method according to the present invention are useful for a peer-to-peer network that distributes and manages content, and are particularly suitable for a network that performs broadband data and a peer-to-peer network that implements various content services. ing.

It is a figure which shows the example of arrangement | positioning of the peer in the one-dimensional hash space of the network system of Embodiment 1 concerning this invention. It is a figure which shows an example of DHT which the peer of a content request origin manages. It is a figure which shows an example of the binary search tree of the peer of a content request source. 5 is a flowchart illustrating an example of a peer search processing procedure according to the first embodiment. It is a figure which shows the concept of the peer search process of Embodiment 1 in the case of h = 31. It is a figure which shows the concept of the peer search process of Embodiment 1 in the case of h = 2. 10 is a flowchart illustrating an example of a peer search processing procedure according to the second embodiment. It is a figure which shows the concept of the peer search process of Embodiment 2 in case of h = 31. It is a figure which shows the concept of the peer search process of Embodiment 2 in case of h = 2. FIG. 11 is a diagram illustrating an example of arrangement of peers in the one-dimensional hash space of the network system according to the third embodiment. FIG. 22 is a diagram illustrating an example of a finger table of a transmission source peer according to the third embodiment. FIG. 20 is a diagram illustrating an example of a binary search tree according to the third embodiment. FIG. 20 is a diagram illustrating an example of a binary search tree according to the third embodiment. FIG. 11 is a diagram illustrating an example of parallel processing timing when the peer search process according to the third embodiment is used. It is a figure which shows an example of the parallel processing timing at the time of using the conventional peer search process. It is a figure which shows the function structural example of the search processing apparatus which implement | achieves the peer search process of this Embodiment by hardware. It is a figure which shows an example of the process at the time of pipelining.

Explanation of symbols

1, 1a One-dimensional hash space 2 Peer 3 Finger peer 4 Comparison circuit 5 Memory 6 Memory bus

Claims (7)

When peers that make up a peer-to-peer network that uses a distributed hash table method to distribute and manage content manage data transfer destination peers determined based on the hash value assigned to each device as transfer destination peer information A peer search method for searching a content request transmission destination or a peer to which the content request is transferred from peers included in the transfer destination peer information,
A search tree generation step of generating a search tree in which a branch condition is defined using a transfer destination peer hash value that is a hash value of a peer included in the transfer destination peer information;
A determination step of determining whether or not a hash value of a content key corresponding to the content request satisfies the branch condition in order from the top of the search tree;
Among the branch conditions determined to satisfy the branch condition, a transfer destination peer hash value corresponding to the branch condition located at the bottom of the search tree is obtained, and a peer corresponding to the transfer destination peer hash value is determined as a search result. A peer determination step,
A peer search method characterized by comprising:   When the peer that manages the content or the peer that has received the content searches for a content transfer peer that is a transfer destination peer to transmit the content to the request source, the content is determined by the determination step and the peer determination step. The peer search method according to claim 1, further comprising searching for a forwarding peer.   2. The method according to claim 1, wherein if the branching condition is never satisfied in the determination step, a peer whose hash value satisfies a specific condition among the peers included in the forwarding destination peer information is set as a search result. 3. The peer search method according to 1 or 2.   Prior to the execution of the determination step, “determining that the hash value of the content key corresponding to the content request satisfies the branch condition at the peer that satisfies the specific condition among the peers included in the transfer destination peer information. The initial state is set, and if there is no branch condition determined to satisfy the branch condition in the determination step, the peer determination step uses the initial state as the determination result of the determination step. The peer search method according to claim 1 or 2, characterized in that:   In the search tree generation step, when the number of peers included in the transfer destination peer information is smaller than the maximum number of peers included in the transfer destination peer information, the maximum search number and the branch condition are added to the generated search tree. The peers included in the forwarding destination peer information are added with one or more peers whose hash values satisfy a specific condition so that the number of the same is equal and a balanced tree. 5. The peer search method according to any one of 4 above.   The peer search method according to any one of claims 1 to 5, wherein each of the determination processes performed in order from the upper stage of the search tree in the determination step is performed by parallel processing. Configures a peer-to-peer network that distributes and manages content using a distributed hash table method, manages each data transfer destination communication device determined based on the hash value assigned to the device as transfer destination peer information, and requests content A communication device that searches a peer included in the transfer destination peer information for a communication device of a transmission destination of the content request or a transfer destination of the content request,
Storage means for storing a search tree in which a branch condition is defined using a transfer destination peer hash value that is a hash value of a peer included in the transfer destination peer information;
The search tree is read from the storage means, and it is determined whether or not the hash value of the content key corresponding to the content request satisfies the branch condition in order from the top of the search tree, and is determined to satisfy the branch condition. Determining means for obtaining a transfer destination peer hash value corresponding to the branch condition located at the bottom of the search tree among the branch conditions, and determining a communication device corresponding to the transfer destination peer hash value as a search result;
A communication apparatus comprising:

Images