JP2009020757A - Data registration apparatus, data registration method and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a data registration apparatus capable of efficiently writing data from a primary data group to a DHT. <P>SOLUTION: In the data registration apparatus 1, each DHT node 3 dispersively stores data made to correspond to a key, on the basis of a hash value of the key. In the data registration apparatus 1, an input part 15 inputs a series of registration data including a key and primary data which is a data source from a primary data table 2, a hash sort part 12 calculates a hash value of the key in each registration data of the series and sorts the order of registration data in the series on the basis of the hash value so that registration data to be stored in the same DHT node 3 are continued, a filter part 13 deletes registration data other than that to be registered again out of the registration data of the series, a writing part 14 writes respective registration data of the series in the corresponding DHT node 3 in accordance with the sorting order. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a data registration apparatus, a data registration method, and a program for registering data in a distributed hash table.

  In a large-scale distributed storage system represented by a distributed hash table (DHT) such as the Chord method (Non-patent document 1) and the Tapestry method (Non-patent document 2), the data handled on the storage system is in soft state. (Non-Patent Document 3). That is, the retention of data on the storage system is guaranteed only for a certain period of time, and the data source (clients existing in a distributed manner) that owns the primary data needs to rewrite as necessary.

When all primary data is held on the DHT in a soft state, if the DHT system scale increases and the amount of data increases, the state to be maintained and the communication for refresh increase in proportion to the amount of data. Do not scale. On the other hand, if data retention is left to the DHT side without using the soft state, data may be lost due to a failure of the DHT node or the like. Furthermore, when performing backup or migration from DHT, it is necessary to store the amount of data expected by DHT. Therefore, since it takes too much time to perform linear backup in a straightforward manner, backup or migration to DHT is not possible. Although it is natural to do this, there has been no way to efficiently copy the contents of one DHT to the other DHT.
I. Stoica, R. Morris, D. Karger, MF Kaashoek, and H. Balakrishnan. Chord: A scalable peer-to-peer lookup service for internet applications.In Proceedings of ACM SIGCOMM, August 2001. Ben Y. Zhao, John D. Kubiatowicz, and Anthony D. Joseph. Tapestry: An infrastructure for fault-tolerant wide-area location and routing. Technical Report UCB // CSD-01-1141, UCBerkeley, April 2000. Sean Rhea, Brighten Godfrey, Brad Karp, John Kubiatowicz, Sylvia Ratnasamy, Scott Shenker, Ion Stoica, and Harlan Yu.OpenDHT: A public DHT service and its uses.In Proceedings of ACM SIGCOMM 2005, August 2005.

  Conventionally, there has been no effective method for writing data from the primary data group to the DHT or writing data from the first DHT to the second DHT.

  The present invention has been made in consideration of the above circumstances, and is capable of efficiently writing data from the primary data group to the DHT or writing data from the first DHT to the second DHT. It is an object to provide a registration device, a data registration method, and a program.

  The present invention provides a data registration apparatus for registering data in a distributed hash table including a plurality of distributed hash node apparatuses that distribute and store data associated with key information based on a hash value of the key information. And means for inputting a series of registration information including primary data that is the source of the data, and hash means for calculating a hash value of the key information related to the registration information for each registration information of the inputted series And the order of the registration information in the inputted sequence is based on the hash value of the key information related to each registration information so that the registration information to be stored in the same distributed hash node device is the sequential order. Sorting means for sorting the order of registration information in the series, and targeting all or part of the registration information included in the sorted series Te, the writing of the registration information to be stored the registration information the distributed hash node apparatus, characterized by comprising a writing means for performing in accordance with the sort order.

  According to the present invention, it is possible to efficiently write data from the primary data group to the DHT or write data from the first DHT to the second DHT.

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

(First embodiment)
FIG. 1 shows a configuration example of a distributed hash table system according to an embodiment of the present invention.

  In FIG. 1, 1 is a data registration device, 2 is a primary data table, 3 is a DHT node, and 8 is a network.

  A plurality of DHT nodes 3 cooperate to form one distributed hash table 30. The number of DHT nodes constituting the distributed hash table 30 is arbitrary. Further, this number can be dynamically changed by joining / leaving the DHT node.

  Each DHT node 3 associates key information (key) and data (data: 1 or a plurality of values) as illustrated in FIG. 2 with respect to key information that provides a hash value corresponding to the hash value that it is in charge of. In response to the inquiry message with the key information specified, which is sent by the client device (not shown), the hash value of the key information is obtained in response to the inquiry message. If the hash value is in charge of itself, a response message including data stored in correspondence with the key information is returned to the client device. If it is not the hash value that it is in charge of, for example, the inquiry message is transferred to another DHT node 3, or a combination of an IP address and a port number or a URL is returned as a reference to a more appropriate DHT node 3. .

  For the DHT node 3, management of information on the own node, participation / leaving control, management of information on adjacent nodes, database management, search for a DHT node in charge of an arbitrary hash value in the hash space to which the own node belongs The procedure for the inquiry message from the client device may basically use a conventional method.

  As illustrated in FIG. 3, each primary data table 2 of the primary data table group 20 includes various primary data (values) that are sources (data sources) of data to be registered in the distributed hash table 30. The key information (key) is held in association with it. The number of primary data tables 2 is arbitrary. Note that primary data having the same key information may be held across a plurality of primary data tables 2.

  The primary data table group 20 may have any configuration. For example, each primary data table 2 of the primary data table group 20 may be a normal independent database, or a plurality of primary data tables. 2 may constitute another distributed hash table 30.

  The data registration device 1 registers the primary data held in the primary data table 2 in the DHT node 3 that is responsible for the primary data (that is, each primary data is a DHT that is responsible for the hash value of the corresponding key information). Registered in node 3). Here, one data registration device 1 is provided corresponding to each primary data table 2. The pair of data registration devices 1 and the primary data table 2 may be devices independent of each other or may be integrated devices.

  A plurality of DHT nodes 3 constituting one distributed hash table 30 can communicate with each other via the network 8. Each data registration device 1 can communicate with an arbitrary DHT node 3 via the network 8. The network 8 may be any network. Here, a case where the network 8 is the Internet will be described as an example.

  The distributed hash table system illustrated in FIG. 1 is a service configuration used for, for example, traceability applications (for example, literature “Tetsu Ozaki, Yusuke Doi, Shiro Wakayama. Product traceability system that can be expanded smoothly from a small scale. Toshiba Review, Vol. 60, No.8, pp.27-31, 2005.)). Hereinafter, a case where the present distributed hash table system is used for constructing an individual product traceability system will be described as an example.

  When applied to an individual product traceability system, for example, each primary data table 2 of the primary data table group 20 includes “information related to the individual product (for example, quality information) as primary data for each individual product. The server information to be provided (for example, URL) ”is held in association with“ identification information unique to the individual product (hereinafter referred to as tag ID ”) as the key information, Each DHT node 3 of the distributed hash table 30 holds the URL (one or more) as secondary data in association with a tag ID in an index table used for lookup or the like. In this case, the URL or the like may indicate a node having the primary data table 2 in FIG. 1 or may indicate another server device.

  FIG. 4 shows GID-96 (EPCglobal. EPC tag data standards version 1.3. EPCglobal Ratified Specification, March 2006) in EPCglobal as an example of a tag ID assigned to each individual product. In this case, the tag ID consists of three layers of company ID, product type ID, and individual product ID. In this case, the DHT is provided for each company or for each company and product type, but is not limited thereto.

  In such an individual item traceability system, for example, a user (for example, a consumer) who wants to acquire and view information on a certain item (individual item) specifies a tag ID of the individual item from a client device (not shown). The inquiry message is transmitted to the distributed hash table 30 (any one of the DHT nodes 3), a list of URLs and the like is acquired from the distributed hash table 30, and then a predetermined server device based on the acquired URL and the like. By transmitting an inquiry message to the device, an operation of acquiring specific information (for example, temperature management history) regarding the individual product is performed.

  Here, the distributed hash table 30 may lose a part of the index table (tag ID: [URL,...) Due to a node failure or the like. At that time, a part of the data held by the specific DHT node 3 is broken, or the specific DHT node 3 breaks down (disengages), and all the data held by the DHT node 3 is lost. There are cases.

  In the former case, the broken data can be recovered by re-registering with the DHT node 3.

  In the latter case, the data held by the failed DHT node 3 can be recovered by re-registering with another DHT node 3. It is desirable to determine in advance which DHT node 3 is to be used for recovery when which DHT node 3 fails. As an example, consider a case where a hash space is formed as a closed number line with both ends connected in a hash space formed of a set of hash values, such as Chord (for example, a hash value ( (For example, the hash value of the address of the DHT node) is assigned, and the range (upper limit value and lower limit value) of the hash value assigned to each DHT node 3 is determined based on the hash value assigned to each DHT node 3) In this case, for example, when a certain DHT node 3 fails, the data held by the DHT node 3 is converted into a node (successor) that is closest to the DHT node 3 in the hash value increasing direction in the hash space. ) Or a node (predecessor) closest to the DHT node 3 in the hash value decreasing direction in the hash space. ) It is, after they missing data can be determined in advance and retain also.

As a hash function for calculating a hash value, for example, SHA-1 or MD5 can be adopted as a function that has a sufficiently long number of bits and is uniformly assigned. The hash space is represented by an m-bit integer of 0 to 2 ^m −1 with m = 160 for SHA-1 and m = 128 for MD5.

  Further, as a method for determining a hash value to be associated with each DHT node 3, for example, a method using a hash value obtained by applying a hash function to an address (for example, an IP address) of the DHT node 3 may be adopted. Alternatively, other methods may be adopted.

  The data registration device 1 of the present embodiment detects a region (failure region) in which the above data is lost and restores it.

  Hereinafter, the data registration device 1 of the present embodiment will be described in detail.

  FIG. 5 shows a configuration example of the data registration device 1 according to the present embodiment. As shown in FIG. 4, the data registration device 1 includes an instruction receiving unit 11, a hash / sort unit 12, a filter unit 13, a writing unit 14, and an input unit 15.

  The instruction receiving unit 11 transmits information indicating a failure area in the target distributed hash table 30 (failure area information), a hash algorithm to be used in the distributed hash table 30, and the same DHT node 3 via the network 8. The registration instruction message including the sorting algorithm for sorting the hash values to be consecutive is received and accumulated.

  As for the above hash algorithm and sort algorithm, for example, a method for expressing a combination of parameters in advance may be defined, and the parameters may be described in the registration instruction message, or Java (registered trademark) or the like may be used. You may make it use the dynamic object reading means in a platform.

  The hash algorithm and sort algorithm may be set in each data registration device 1 without being included in the registration instruction message.

  The failure area information is, for example, a list of hash values (h (key)) of key information (key) of failed data. The format of the list may be any format. For example, the list may be described by a method that lists all hash values, or may be described by a method that lists pairs of hash value start points and end points. They may be used in combination.

  Note that various variations are possible for the method of enabling the instruction receiving unit 11 to receive the registration instruction message.

  For example, the distributed hash table 30 as a whole is assumed to share a table called a failure area list for recording a failure area (for example, one management node (not shown) other than the DHT node 3 is provided, and the failure area list is provided in this table). Each DHT node 3 periodically checks whether or not its own device has a failure area, and if it has a failure area, the failure area entry (failure area start point, (End point of failure area) shall be added.

  For example, when a certain DHT node 3 breaks down due to a failure, the certain DHT node 3 cannot add an entry of the failure area, but instead of the certain DHT node 3 After the DHT node 3 (for example, a successor or predecessor) to be detected detects that the certain DHT node 3 has left, all the data to be held by the certain DHT node 3 has been lost. An entry can be added. Also, mainly for the purpose of improving fault tolerance, the same contents (mirror) as the data handled by a certain DHT node 3 are always transferred to the certain DHT node 3 when the certain DHT node 3 fails and leaves. When a DHT node 3 to be replaced by a DHT node 3 (for example, a successor or a predecessor) holds, after the DHT node 3 to be replaced detects the detachment of the certain DHT node 3, If there is a failure area in the data handled by a certain DHT node 3, an entry for the failure area can be added to the failure area. Similarly to a backup instruction based on the time described later, time information may be added to the entry of the failure area in order to perform data restoration based on the last update time of the mirror. In this case, it means that there is a possibility that data newer than the time recorded in the entry is lost among the data included in the failure area.

  Then, the data registration device 1 periodically transmits an inquiry message regarding the failure area to the management node, and the management node that has received the inquiry includes the failure area list in the registration instruction message and registers the data. You may make it return to the apparatus 1. FIG.

  Alternatively, the management node may periodically broadcast a registration instruction message including the failure area list to all the data registration devices 1.

  In the above description, one management node other than the DHT node 3 is provided. However, a plurality of management nodes may be provided, or one or more DHT nodes 3 may be provided instead of or in addition to the management node. The management node may be assigned a role. In addition, when a plurality of management nodes or the like holding a failure area list are provided, the data registration apparatus 1 may transmit an inquiry message regarding the failure area to any one of them. In addition, when the management node periodically transmits a registration instruction message including the failure area list to the data registration apparatus 1, when a plurality of management nodes are provided, the nodes are instructed to register. You may share the data registration apparatus 1 which should transmit a message.

  In addition, when a plurality of management nodes are provided, it is possible to share the DHT node 3 that is the target of the failure area list with each management node.

  By the way, it is preferable that each entry in the failure area list expires at time t in order to prevent the information of the repaired data from remaining indefinitely. However, the individual data registration device 1 must restore the data corresponding to the tag ID having the data itself if it is lost. Therefore, the data registration device 1 expires before the restoration and fails to restore. In order to avoid this, it is preferable to make an inquiry about the failure area to the management node or the like at an interval shorter than the time t so as to reliably acquire the failure area list and update the failed data. In addition, when the management node or the like periodically transmits a registration instruction message including the failure area list to the data registration apparatus 1, the registration instruction message is transmitted at an interval shorter than the time t. Is preferred.

  The hash sort unit 12 follows the continuity of the hash value space of the distributed hash table 30 in order to perform efficient writing in consideration of the continuity of the hash value space of the distributed hash table 30 to be written. Sort data from the primary data source.

  The hash sort unit 12 inputs, through the input unit 15, all the entries (key, value) held by the corresponding primary data table 2 as a sequence of (key, value) in an arbitrary order. For each entry of the (key, value) sequence, a hash value h (key) of the key information key is calculated according to the instructed hash algorithm to obtain (h (key), key, value), All (h (key), key, value) are sorted according to the indicated sorting algorithm to create a sorted (h (key), key, value) column.

  For example, each of the plurality of DHT nodes 3 stores data related to key information that gives a hash value included in a range of consecutive hash values assigned to the own device from the hash value space used in the distributed hash table 30. If so, the sorting algorithm sorts so that writing is performed according to the order of the hash value space used in the distributed hash table 30.

  As a specific example, for example, in the distributed hash table 30 according to Chord, the hash algorithm may be MD5 or SHA-1, and the sort algorithm may be ascending order or descending order, ascending order of random start, or descending order of random start.

  Note that the random start sort algorithm uses a randomly selected hash value so that the DHT node 3 that starts writing is random so that data writing does not concentrate on a specific DHT node 3. It sorts so that it becomes. For example, in a distributed hash table algorithm such as Chord, the space of hash values is expressed by a closed number line, and only the context is defined. Therefore, all values can be written without omission by going around the hash value space no matter where the number line starts writing.

  The filter unit 13 targets the sorted (h (key), key, value) column given from the hash sort unit 12 to the writing unit 14 based on the designated “failure area information”. , Filtering. That is, since the hash value of the data to be re-registered is described in the failure area information, data other than the data to be re-registered is deleted from the sorted (h (key), key, value) column. (That is, h (key) is not described in the failure area information (h (key), key, value) may be deleted).

  The filter unit 13 may be provided between the hash / sort unit 12 and the writing unit 14.

  The writing unit 14 writes data in chunks (bulk) while determining the continuity of the hash value space of the distributed hash table 30 to be written and the area in charge of the DHT node 3.

  The writing unit 14 sorts and filters each registered information (key, value) in the column of (h (key), key, value) via the network 8 from the head (or from a random position). In order, the data is written in the distributed hash table 30 (corresponding DHT node 3).

  The reason for starting from a random position is to prevent data writing from being concentrated on a specific DHT node 3. In addition, a certain restriction may be imposed on the writing speed (such as the time interval for changing the write destination node) so as not to cause congestion near the slow DHT node (for example, a random waiting time is provided). .

  FIG. 6 shows an example of the processing procedure of the data registration device 1.

  The data registration device 1 includes failure area information (for example, failure area list) (in the target distributed hash table 30), a hash algorithm, and a sort algorithm included in the registration instruction message received during a certain period. Extraction and transmission of these to the instruction receiving unit 11 trigger the restoration operation (step S1).

  If new failure area information is not detected in step S2 (including the case where the registration instruction message is not received for a certain period), the subsequent processing is skipped and the process returns to step S1.

  If new failure area information is detected in step S2 (for example, there is failure area information newly added to the failure area list), that area (failure area start point, failure area end point) Is set in the filter unit 13 and the hash algorithm and sort algorithm are set in the hash sort unit 12, and then data restoration is started.

  First, the hash sort unit 12 receives a sequence (in any order) of data in (key, value) format to be written to the target distributed hash table 30 from the primary data table 2 that is a primary data source, It is set after calculating h (key) using the set hash algorithm h (step S3) and converting the (key, value) column to the (h (key), key, value) column. Sorting is performed according to the sorting algorithm (step S4). The sorted columns of (h (key), key, value) are sorted so that hash values assigned to the same DHT node 3 are continuous (for example, in the case of chord, h (key) is hashed). Since it is sorted so as to be continuous in the value space), it can be written efficiently by the writing unit 14.

  Next, the writing unit 14 inquires the filter unit 13 about data to be written, and the filtering unit 13 performs filtering based on the set failure area information (that is, sorted (h (key), key, The entry that does not include h (key) in the failure area information is deleted from the column of value), and the column of data to be written (h (key), key, value) is returned to the writing unit 14 (step S5). The writing unit 14 performs writing for each (h (key), key, value) in the column of the filtered and sorted data (h (key), key, value) (step S6). Note that (key, value) sorted by h (key) may be returned from the filter unit 13 to the writing unit 14.

  FIG. 7 shows an example of the write processing procedure in step S6.

  First, one appropriate DHT node 3 is selected and set from the target distributed hash table 30 (step S11).

  Next, among the data (key, value) that has not yet been written in the column of the filtered and sorted data (h (key), key, value) (or (key, value)), One with the earliest sort order is selected, and a write message including (key, value) is transmitted to the currently set DHT node 3, thereby attempting to write the data (key, value) (step) S12).

  Here, the DHT node 3 that has received the write message including (key, value) to be written from the data registration device 1 has the hash value h (key) calculated from the key as the device in charge. The (key, value) is registered in its own device. If h (key) is not in charge of the own device, the DHT node 3 or another DHT node 3 or the like is a DHT node 3 (DHT in charge) closer to the DHT node 3 in charge of the hash value h (key). Node 3 is resolved) and a redirect message indicating the closer DHT node 3 is returned to the data registration device 1.

  For example, each DHT node 3 stores a DHT node (subsequent DHT node) located immediately after a series of hash values handled by the own device, and the hash value h (key) is in the assigned range of the own device. The DHT node 3 that has received the write message including no (key, value) may return the information including the information of the subsequent DHT node in the redirect message. In this case, if the subsequent DHT node is correct, the next writing is successful, but if it is not correct, redirect messages are successively returned until the writing is successful. Of course, the DHT node 3 that has received the write message that includes the (key, value) whose hash value h (key) is not in the scope of its own device inquires the other DHT node 3 about the correct DHT node that is responsible for the hash value. The DHT node 3 that has received the write message or the DHT node 3 that has received the inquiry message, etc. after resolving the correct DHT node by transmitting a message or the like, transmits information indicating the correct DHT node as a redirect message. You may make it reply including it, and you may use another method.

  If redirection occurs (step S13), the correct DHT node 3 to which the data returned from the DHT node 3 is to be written is reset (step S11), and the data is written again (step S12).

  If redirection does not occur and writing is successful (step S13), an attempt is made to write the next data in the sort order (step S12).

  If there is a plurality of data to be written to the same DHT node 3 by the above sorting, they are sorted in a sequential order, so that the writing in step S12 is continuous until the DHT node 3 to be written next changes. And will be successful.

  The above process is repeated while there is the next data in step S14, and the process ends when all the data is written.

  In the example of FIGS. 6 and 7, after all the data is filtered in step S5, writing is performed in step S6. However, when one data is selected in step S12 of FIG. This data is filtered (that is, whether or not the data is to be written based on the failure area information). If the data is to be written, the data is written. If the data is not to be written, the process proceeds to step S14. This may be repeated.

  Hereinafter, a more specific example of the writing procedure will be described.

  For example, in the case of Chord, it becomes a closed 160-bit one-dimensional hash value space defined by SHA1, and a certain area on the one-dimensional space is managed by a certain DHT node. If a data string that is not sorted is written, a DHT node search based on the Chord algorithm is executed for each write and the efficiency deteriorates. In this embodiment, the hash sort unit 12 distributes the write destination. By writing data in a sequence of data in accordance with the continuity of the hash value space of the hash table 30, it becomes possible to collect writes to the area handled by a specific DHT node, thereby reducing the search cost. Similarly, when writing to the range is completed and a destination for writing the next data entry is searched, it is only necessary to move the writing destination based on the continuity of the hash value space.

  In the above, each DHT node 3 constituting the distributed hash table 30 has key information (key) and data (data: 1 or multiple values) for key information that provides a hash value corresponding to the hash value that it is responsible for. ), And a response message including data stored corresponding to the key information is returned in response to the inquiry message specifying the key information from the client device. Instead, each DHT node 3 holds the hash value (h (key)) of the key information and the data (data: 1 or multiple values) in association with each other, and specifies the hash value of the key information from the client device In response to the inquiry message, the response message including the stored data corresponding to the hash value of the key information. It may be the one to return the. In this case, in each embodiment, after calculating the hash value h (key) from the key information (key), the data registration device 1 may handle (h (key), value) (key information ( (key) does not have to be handled). In addition, (h (key), value) may be added to the write message transmitted to the DHT node 3 instead of (key, value). This also applies to the second and third embodiments.

  FIG. 8 shows an example of algorithm for writing by pseudocode. In this example algorithm, filtering in step S4 in FIG. 6 and writing in step S5 are integrated. In FIG. 6, a sentence starting with // is a comment sentence.

  In this example algorithm, TupleStream is a Tuple iterator, and Tuple is a combination of hkey and value. hkey is h (key) corresponding to (key, value) on the primary data side, corresponding to h (key). Sorted_source is sorted (before being filtered) (h (key), value).

  The TupleFilter is a filter based on failure area information for (h (key), value).

  The DHTNode is a class representing the DHT node 3 of the target distributed hash table 30. nptr is a pointer to an instance of DHTNode indicating one node of the DHT node 3, and includes a URL or an IP address.

  nptr. put (t.hkey, t.value) writes one (h (key), value) to the DHT node 3 indicated by nptr. Here, the case of transmitting h (key) is described, but when transmitting the key, t. hkey to t. The key may be used.

  The DHT node 3 returns a reference to a DHTNode that is considered to be more correct together with the RedirectException for a write request outside the range handled by the own device. The catch receives this. nptr = re. redirected_to sets an instance of DHTNode generated from the URL or IP address of the returned DHT node 3 to nptr.

  Here, the discovery and advertisement of the failure area information (failure area list) of the target distributed hash table 30 may be different depending on the DHT algorithm. For example, in the case of Chord, node detachment / failure can be found by fluctuation of the predecessor value (indicating the previous node) in the DHT hash value space. If the DHT node does not have a mirror of the area owned by the detached node, the area information of the detached DHT node is added to the failure area list information of the DHT node along with the time, and the updated failure area list Advertise As an example, the advertisement of the failure area list may be performed for each DHT node that is a destination of a finger that is a short-circuit path on a successor and a chord (indicating a later node). The DHT node that has received the failure area list advertisement compares the failure area list with its own failure area list, merges new data into its own failure area list only when new data exists, and advertises similarly. . For this comparison, an ID based on a pseudo-random number, time, or the like may be attached to the generated failure area list.

  By combining the failure area list discovery, advertisement, and data registration device, the primary data table group 20 and the index DHT can quickly and inexpensively recover from data loss due to a DHT failure or the like.

  According to the present embodiment, it is possible to efficiently perform restoration from the primary data source for data loss on the distributed hash table.

(Second Embodiment)
In the following, the second embodiment will be described focusing on the differences from the first embodiment.

  FIG. 9 shows a configuration example of the distributed hash table system according to the present embodiment.

  In the present embodiment, the primary data table group 20 of the first embodiment is a first distributed hash table 40 (the primary table 2 is a DHT node 4), and the distributed hash table 30 of the first embodiment is As the second distributed hash table 30, the second distributed hash table 30 is used as a backup of the first distributed hash table 40.

  Note that a plurality of backups may be provided (a plurality of second distributed hash tables 30 may be provided). In this case, the backup may be performed independently for each system.

  FIG. 10 shows a configuration example of the data registration device 1 according to the present embodiment. This configuration example is basically the same as that shown in FIG.

  In this embodiment, when performing backup of the first distributed hash table 30 on the second distributed hash table 40, the data registration apparatus 1 performs differential backup continuously (for example, every time a certain period elapses). Will be described as an example. Note that this differential backup is effective, for example, when it is performed in a distributed hash table or a time zone when the network load is low.

  For example, the control device 7 provided for the first distributed hash table 40 and the second distributed hash table 30 backs up each data registration device 1 (the instruction receiving unit 11) simultaneously or sequentially. An instruction may be issued (individual data registration apparatus 1 receives a backup instruction at an appropriate interval), or the control apparatus 7 is not provided, and the individual data registration apparatus 1 voluntarily In addition, a backup instruction may be issued to the instruction receiving unit 11 continuously (for example, every elapse of a certain period of time).

  The hash algorithm and sort algorithm for the second distributed hash table 30 are basically the same as those in the first embodiment. For example, the control device 7 uses the hash algorithm and sort algorithm as a backup instruction message. It may be given to each data registration device 1 or may be set in each data registration device 1 without being included in the backup instruction message.

  In the present embodiment, for example, the control device 7 (or the data registration device 1 itself) indicates the “previous backup instruction time” to the instruction receiving unit 11 of the data registration device 1, so that time A method may be considered in which only the (key, value) data entry updated thereafter is updated. At this time, the function of recording the registration time of each (key, value) data entry in the DHT node 4 as the primary data source and the clock of the DHT node 4 as the individual primary data source by a function such as NTP. It is desirable to provide a function for aligning. For example, the registration time reg_t is added to the data entry (key, value) to obtain (key, value, reg_t), and the data entry registered at the time after the “last backup instruction time” (that is, “ The filtering unit 13 may perform filtering with reference to the registration time reg_t so that only the data entry having the registration time reg_t that is a time after the “last backup instruction time” is registered.

  When only one backup is provided, it is possible to selectively update only the data that has not been backed up more easily by a management method such as dirty bit. For example, the instruction receiving unit 11 sets the filter unit 13 so as to register only the data entry in which the dirty bit is set, and the DHT node 4 that is the primary data source starts from the data entry that has been backed up. The bit should be canceled.

  FIG. 11 shows an example of a processing procedure of the data registration device 1 of the present embodiment.

  The difference between step S21 and step S2 (FIG. 6) is that the process starts when the instruction receiving unit 11 receives a backup instruction (including, for example, “previous backup instruction time”) in step S21. .

  Steps S22 and S23 (hash calculation and sorting) are the same as steps S3 and S4 in FIG.

  The difference between step S24 (filtering) and step S5 (FIG. 6) is that filtering is performed so that differential backup based on the previous backup instruction time or dirty bit is performed.

  Step S25 (write) is the same as step S6 in FIG. 6 and FIG.

  Further, the procedure example of FIG. 8 shown in the first embodiment can also be used in the present embodiment except that the contents of filtering are different.

  Also in this embodiment, the data registration apparatus 1 provided on the distributed hash table 40 side may also restore the failure area in the distributed hash table 30 as in the first embodiment.

  Further, the data registration device 1 may be provided on the distributed hash table 30 side, and the failure area in the distributed hash table 40 may be restored as in the first embodiment.

  According to this embodiment, differential backup between distributed hash tables can be performed efficiently.

(Third embodiment)
In the following, the third embodiment will be described focusing on the differences from the first and second embodiments.

  A configuration example of the distributed hash table system according to the present embodiment is the same as that shown in FIG.

  A configuration example of the data registration device 1 according to the present embodiment is the same as that shown in FIG.

  In the second embodiment, the differential backup from the first distributed hash table 40 to the second distributed hash table 30 is performed. However, in the present embodiment, the first distributed hash table 40 is changed to the second distributed hash table. 30 (copying) to 30. Note that copying may be performed for any purpose, for example, backup or replacement of a distributed hash table system.

  Note that, for example, the control device 7 provided for the first distributed hash table 40 and / or the second distributed hash table 30 has the data receiving device 1 (the instruction receiving unit 11 thereof) in the first distributed hash table 40. At the same time or sequentially, an entry point (for example, an IP address and a port number) to the second distributed hash table 30 is added to the copy instruction to the second distributed hash table 30 and transmitted. The hash algorithm and sort algorithm for the second distributed hash table 30 are basically the same as those in the first and second embodiments.

  Each data registration device 1 in the first distributed hash table 40 is similar to the first and second embodiments in that the sort algorithm is set in the hash / sort unit 12, but here the corresponding DHT node 4 In order to back up all data stored in the file, the write area is not limited by setting empty data (null) in the filter unit 13.

  The hash sort unit 12 receives a column of data (key, value) as an input and uses the specified hash algorithm and sort algorithm of the second distributed hash table 30 to sort the data sequence (h (key) , Key, value) to the writing unit 14.

  The writing unit 14 starts writing the given data from the beginning (or from a random position). The writing unit 14 writes the data as a block to a certain DHT node 3 of the second distributed hash table 30. When the data to be written is out of the area in charge of the node, the write destination is changed to an appropriate adjacent node in the space defined by the second distributed hash table 30 by means such as error notification. The reason for starting from a random position is to prevent a situation in which each data registration device 1 in the first distributed hash table 40 writes to a specific DHT node 4 in the second distributed hash table 30. It should be noted that a certain restriction may be imposed on the writing speed (such as the time interval for changing the write destination node) so as not to cause congestion in the vicinity of the slow DHT node (for example, a random waiting time is provided). .

  FIG. 12 shows an example of the processing procedure of the data registration device 1 of this embodiment.

  The difference between step S31 and step S2 (FIG. 6) is that the process starts when the instruction receiving unit 11 receives a copy instruction (including, for example, an IP address and a port number) in step S21.

  Steps S32 and S33 (hash calculation and sorting) are the same as steps S3 and S4 in FIG.

  The difference between step S34 (filtering) and step S5 (FIG. 6) is that the contents of filtering are filtered so that all data is a copy target. In the present embodiment, the filter unit 13 and step S34 can be omitted.

  Step S35 (writing) is the same as step S6 in FIG. 6 and FIG.

  Further, the procedure example of FIG. 8 shown in the first embodiment can also be used in the present embodiment except that the contents of filtering are different.

  In this embodiment, the data registration device 1 provided on the distributed hash table 40 side may also restore the failure area in the distributed hash table 30 as in the first embodiment. Similar to the second embodiment, differential backup may be performed.

  Further, the data registration device 1 may be provided on the distributed hash table 30 side, and the failure area in the distributed hash table 40 may be restored as in the first embodiment.

  According to the present embodiment, data migration can be efficiently performed from the first distributed hash table to the second distributed hash table.

Each of the above functions can be realized even if it is described as software and processed by a computer having an appropriate mechanism.
The present embodiment can also be implemented as a program for causing a computer to execute a predetermined procedure, causing a computer to function as a predetermined means, or causing a computer to realize a predetermined function. In addition, the present invention can be implemented as a computer-readable recording medium on which the program is recorded.

  Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriately combining a plurality of components disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, constituent elements over different embodiments may be appropriately combined.

The figure which shows the structural example of the distributed hash table system which concerns on one Embodiment of this invention. The figure which shows an example of the data which DHT node 3 hold | maintains The figure which shows an example of the data which a primary data table hold | maintains The figure which shows an example of tag ID The figure which shows the structural example of the data registration apparatus which concerns on the same embodiment The flowchart which shows an example of the process sequence of the data registration apparatus which concerns on the embodiment The flowchart which shows an example of the write-in processing procedure of step S6 of the procedure of FIG. Diagram showing an example algorithm for writing The figure which shows the other structural example of the distributed hash table system which concerns on the embodiment The figure which shows the other structural example of the data registration apparatus which concerns on the same embodiment The flowchart which shows the other example of the process sequence of the data registration apparatus which concerns on the embodiment The flowchart which shows another example of the process sequence of the data registration apparatus which concerns on the embodiment

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Data registration apparatus, 2 ... Primary data table, 3, 4 ... DHT node, 7 ... Control apparatus, 8 ... Network, 11 ... Instruction receiving part, 12 ... Hash sort part, 13 ... Filter part, 14 ... Write , 15 ... input unit, 20 ... primary data table group, 30, 40 ... distributed hash table

Claims (15)

In a data registration device for registering data in a distributed hash table composed of a plurality of distributed hash node devices that distribute and store data associated with key information based on a hash value of the key information.
Means for inputting a sequence of registration information including the key information and primary data that is the source of the data;
Hashing means for calculating a hash value of the key information related to the registration information for each input registration information of the series,
Based on the hash value of the key information related to each registration information, the order of the registration information in the inputted sequence is the order in which the registration information to be stored in the same distributed hash node device is continuous. A sorting means for sorting the order of registration information;
Write means for writing the registration information to the distributed hash node device that should store the registration information for all or part of the registration information included in the sorted series according to the sort order. A data registration device characterized by comprising. The writing means includes
Node information indicating a distributed hash node device selected as a writing destination from a plurality of distributed hash node devices in advance is held,
In writing the registration information to be written in the series, first, the registration information was tried to be written using the node information as a write destination, and as a result, the registration information was successfully written. In this case, the process shifts to writing registration information having the next order to be written in the series, while the distributed hash node device to store the registration information is different from the distributed hash node device indicated by the node information. Therefore, when the writing of the registration information fails, the writing of the registration information is tried again after changing the node information to another distributed hash node device. Data registration device.   When the writing means fails to write the registered information, the writing unit should store the registered information from the distributed hash node device indicated by the node information or another distributed hash node device. 3. The data registration device according to claim 2, wherein information indicating a distributed hash node device closer to the distributed hash node device is received, and the node information is changed to information indicating the received correct distributed hash node device. .   4. The writing unit according to claim 1, wherein the writing unit performs the writing for all of the registration information included in the sorted series when a copy command is received from the outside. The data registration device according to item. Prior to the writing means performing the writing, the writing means further comprises selection means for selecting whether to perform the writing for each registration information included in the sorted series,
4. The writing unit according to claim 1, wherein the writing unit performs the writing only on the registration information included in the sorted series that is selected by the selection unit to perform the writing. The data registration device according to any one of the above. Means further comprising means for receiving a message indicating list information of hash values of key information relating to missing data in the distributed hash table;
The selection means selects, for each piece of registration information included in the sorted series, the writing to be performed when the hash value of the key information related to the registration information is included in the list information. The data registration apparatus according to claim 5, wherein the data registration apparatus is a data registration apparatus.   The selection means, for each registration information included in the sorted series, when the registration information has not yet been written since the content of the primary data related to the registration information has been updated. 6. The data registration apparatus according to claim 5, wherein the data registration is selected to be performed. The registration information includes information indicating the last update time of the primary data related to the registration information,
The said selection means selects the said writing when the information which shows the last update time contained in the said registration information shows the time after the time regarding the last writing, It is characterized by the above-mentioned. 8. The data registration device according to 7.   9. The data according to claim 1, wherein the writing unit provides a random waiting time when a distributed hash node device as a writing destination is changed in the writing. Registration device. Each of the plurality of distributed hash node devices stores data related to key information that gives hash values included in a range of consecutive hash values assigned to the own device from the hash value space used in the distributed hash table. Is,
10. The sorting means according to claim 1, wherein the sorting means sorts the order of registration information in the input sequence in ascending or descending order of hash values of key information related to the registration information. The data registration device described.   11. The writing unit according to claim 1, wherein the writing unit starts the writing of each registration information included in the sorted series from registration information randomly selected from the sorted series. The data registration device according to item 1.   12. The data registration apparatus according to claim 1, wherein the writing unit performs the writing repeatedly at regular intervals. Means for receiving a message indicating hash algorithm and sort algorithm information for the distributed hash table;
The hash means calculates the hash value from the key information according to the hash algorithm,
13. The data registration apparatus according to claim 1, wherein the sorting unit sorts the order of registration information in the series according to the sorting algorithm. In a data registration method of a data registration device for registering data in a distributed hash table composed of a plurality of distributed hash node devices that distribute and store data associated with key information based on a hash value of the key information.
Inputting a sequence of registration information including the key information and primary data that is the source of the data;
Calculating a hash value of the key information related to the registration information for each registration information of the input sequence;
Based on the hash value of the key information related to each registration information, the order of the registration information in the inputted sequence is the order in which the registration information to be stored in the same distributed hash node device is continuous. Sorting the order of registration information;
Writing the registration information to the distributed hash node device that should store the registration information for all or part of the registration information included in the sorted sequence according to the sort order. A data registration method characterized by the above. In a program for causing a computer to function as a data registration device that registers data in a distributed hash table composed of a plurality of distributed hash node devices that distribute and store data associated with key information based on a hash value of the key information ,
Means for inputting a sequence of registration information including the key information and primary data that is the source of the data;
Hashing means for calculating a hash value of the key information related to the registration information for each input registration information of the series,
Based on the hash value of the key information related to each registration information, the order of the registration information in the inputted sequence is the order in which the registration information to be stored in the same distributed hash node device is continuous. A sorting means for sorting the order of registration information;
Write means for writing the registration information to the distributed hash node device that should store the registration information for all or part of the registration information included in the sorted series according to the sort order. A program that causes a computer to function.

Images