JP5558318B2 - Distributed management network system - Google Patents

Description

  The present invention relates to a distributed management system that distributes and manages data to a plurality of nodes in a structured overlay network.

  Structured overlay is a network technology that assigns a unique node ID to each node on a computer network and is constructed in the application layer based on this ID.

  A structured overlay is expected to be a fundamental technology for large-scale autonomous distributed systems because of its excellent fault tolerance and scalability. A typical application example is a distributed hash table (hereinafter referred to as “DHT”). ") Is known.

  DHT is a technique for managing a hash table in a distributed manner by a plurality of nodes. In DHT, the node that manages the data is uniquely determined from the hash value of the data, and thus has a very high search success rate and search efficiency for the target data.

  Although DHT has been put into practical use mainly in the field of file distribution (see Non-Patent Documents 1 and 2), it has recently been applied to VoIP (Voice over IP) systems (see Non-Patent Documents 3 to 7). ).

  Further, a VoIP system in a private network closed in the office / premises can be constructed at low cost by using DHT. However, DHT has been pointed out as one of the problems of delay expansion at the time of data retrieval (hereinafter referred to as “Lookup delay”) due to the difference between the network topology of the application layer and the network topology of the lower layer. (See Patent Document 8).

  In order to eliminate the Lookup delay, there is a technique that considers physical proximity between nodes, and among them, a technique called Proximity Identifier Selection (hereinafter referred to as “PIS”) is known (Non-Patent Document 6, 9-11).

  PIS is a technique for reducing Lookup delay by giving each node a Node-ID that takes into account the physical proximity difference between the nodes (see Non-Patent Document 6). PIS is easy to implement, is independent of the structured overlay routing algorithm used, and does not require periodic message transmission / reception between nodes, and therefore has the advantage of not burdening the network.

  As techniques for applying this PIS, distributed binning (see Non-Patent Document 9), LDHT (Locality-aware Distributed Hash Table) (see Non-Patent Document 10), LPDHT (Locality-Aware and Partition-Space DHT) (Non-Patent Document) 6) and Location-based node IDs (see Non-Patent Document 11) are known.

  In distributed binning, each node measures RTT (Round Trip Time) for Landmarks distributed on the network, and ranks the Landmarks in ascending order of RTT. This order of Landmark is called "bin". In distributed binning, bins are quantified by assigning a level corresponding to RTT to Landmark and used as a Node-ID.

  For example, when the RTT is 0 to 100 ms, when the level is 0, 100 to 200 ms, when the level 1 is 200 ms or more, and when the level 2 is defined, the RTT for a Landmark l1, l2, or l3 is 232 ms, respectively , 51 ms and 117 ms, the bin of this node is “l2, l3, l1”, and the binarized value is “012”, which is used as the prefix of the Node-ID.

  In LDHT, each node acquires its own AS (Autonomous System) number by WHOIS query / response, and uses the acquired AS number as a prefix of Node-ID. In LPPDHT and Location-based node IDs, numbers are mapped to geographic domains and two-dimensional geographic areas, and the mapped numbers are used as prefixes for Node-IDs.

P. Druschel and A. Rowstron, "PAST: A large-scale, persistent peer-to-peer storage utility", HotOS VIII, schuss Elmau, Germany, May 2001. M.J.Freedman, E. Freudenthal, and D. Mazie`res, "Democratizing Content Publication with Coral," Proc. Networked Systems Design and Implementation (NSDI), pp. 239-252, 2004. R. Cox, A. Muthitacharoen, and R. Morris, "Serving DNS using Chord," presented at the 1st Int.Workshop on Peer-to-Peer Systems (IPTPS'02), Cambridge, MA, Mar. 2002. C. Jennings, B. Lowekamp, Ed., E. Rescorla, S. Baset, and H. Schulzrinne, "REsource LOcation And Discovery (RELOAD) Base Protocol", draft-ietf-p2psip-base-08, http: // datatracker.ietf.org/doc/draft-ietf-p2psip-base/, March 2010. Guangyu Shi, Jian Chen, Hao Gong, Lingyuan Fan, Haiqiang Xue, Qingming Lu, and Liang Liang, "SandStone: A DHT Based Carrier Grade Distributed Storage System", Parallel Processing, 2009, ICPP '09, International Conference on, pp 420 -428. Wei Mi, Chunhong Zhang, Xiaofeng Qiu, Lichun Li, Yan Wang, and Yang Ji, "LPDHT: A Locality-Aware and Partitioned-Space Architecture for Peer-to-Peer SIP", Wireless Communications, Networking and Mobile Computing, 2009, WiCom '09. 5th International Conference on. Mosiuoa Tsietsi, Alfredo Terzoli, and George Wells, "Prototyping a P2P SIP User Agent With Support for Multiple Overlays", Pervasive Computing and Communications, 2008, PerCom 2008, Sixth Annual IEEE International Conference on. J. Risson and T. Moors, "Survey of Research towards Robust Peer-to-Peer Networks: Search Methods", RFC4981, September 2007. Sylvia Ratnasamy, Mark Handley, Richard Karp, and Scott Shenker, "Topologically-Aware Overlay Construction and Server Selection", In Proceedings of the INFOCOMM, 2002. Weiyu Wu, Yang Chen, Xinyi Zhang, Xiaohui Shi, Lin Cong, Beixing Deng, and Xing Li, "LDHT: Locality-aware Distributed Hash Tables", Information Networking, 2008, ICOIN 2008, International Conference on. Shuheng Zhou, Gregory R. Ganger and Peter Steenkiste, "Location-based node IDs: enabling explicit locality in DHTs," Carnegie Mellon University, Tech. Rep. CMU-CS-03-171, 2003.

  However, in distributed binning, a large deviation in node density is caused by the bin bins that are quantified, resulting in a short overlay path length between nodes, resulting in a large deviation in the amount of data managed by the node. There was a problem.

  Also, in LDHT, since the AS number is used as it is for the prefix, there is a problem that the deviation in the amount of data managed by the node becomes large as in distributed binning.

  Also, in LPDHT and Location-based node IDs, two-dimensional geographic information is used as it is as a Node-ID prefix, so a small network such as a private network in which nodes exist in a three-dimensional environment such as a building at high density. There was a problem that it was difficult to apply to this.

  The present invention has been made in order to solve such a conventional problem, and can be applied to a small-scale network in a three-dimensional environment, and can be applied to a small-scale network in a three-dimensional environment and can suppress a deviation in the amount of data managed by a node. The purpose is to provide a system.

  The distributed management network system of the present invention includes a plurality of nodes including at least one bootstrap node and connected to the network, the node having a node ID based on a hash value of unique attribute information, and a virtual Based on the position mapped on the circumference according to the node ID and mapped on the virtual circumference and the position of the data mapped on the virtual circumference according to the hash value, the data is In a distributed management network system in which a node to be managed is determined, the bootstrap node overwrites a part of the head of the node ID with a uniquely assigned fixed-length prefix, and is other than the bootstrap node in the node The non-bootstrap node is the closest bootstrap node on the network. The prefix of the node ID is overwritten with the prefix of the node ID with a prefix having the same variable length, and the bit length of the prefix of the non-bootstrap node is set to the non-bootstrap node on the network. Between the position at which the nearest bootstrap node is mapped on the virtual circumference and the position of the bootstrap node mapped to sandwich the non-bootstrap node with the bootstrap node on the virtual circumference. It has a configuration that is determined so as to become shorter as the distance becomes longer.

  With this configuration, the distributed management network system of the present invention can perform mapping so that the nodes are not concentrated on the virtual circumference, so that it is possible to suppress a bias in the amount of data managed by the nodes.

  In addition, the distributed management network system of the present invention can be applied to a small-scale network in a three-dimensional environment in order to determine a node for managing data after mapping nodes on a virtual circumference.

  The bootstrap node has a prefix obtained by inverting the upper and lower bit positions of a value obtained by subtracting 1 from the bootstrap node number assigned consecutively from 1 in the network. A part may be overwritten.

  With this configuration, the distributed management network system of the present invention can perform mapping so that bootstrap nodes are not concentrated on the virtual circumference.

  Further, the lengths of the prefixes of the non-bootstrap nodes may be determined to be all equal when the bootstrap nodes are mapped so as to substantially divide the virtual circumference. Good.

  With this configuration, the distributed management network system of the present invention can map non-bootstrap nodes so as to be substantially equally divided between bootstrap nodes on the virtual circumference.

  Further, when the bootstrap node is not mapped so as to substantially divide the virtual circumference, the non-map mapped to a region where the length between the bootstrap nodes on the virtual circumference is long. The length of the prefix of the bootstrap node is determined to be 1 bit shorter than the bit length of the prefix of the non-bootstrap node mapped in the region where the length between the bootstrap nodes is short on the virtual circumference. It may be.

  With this configuration, the distributed management network system of the present invention can perform mapping so that non-bootstrap nodes are not concentrated between bootstrap nodes on a virtual circumference.

  The present invention can be applied to a small-scale network in a three-dimensional environment, and can provide a distributed management network system that can suppress a deviation in the amount of data managed by a node.

1 is a block diagram showing a configuration of a distributed management network system according to an embodiment of the present invention. FIG. It is a hardware block diagram of the node which comprises the distributed management network system which concerns on one embodiment of this invention. It is a conceptual diagram which shows the prefix which the bootstrap node which comprises the distributed management network system which concerns on one embodiment of this invention produces | generates. It is a conceptual diagram for demonstrating the mapping of the node with respect to the virtual circumference in the distributed management network system which concerns on one embodiment of this invention, (a) shows the example which mapped the bootstrap node on the virtual circumference (B) shows an example in which nodes are mapped on the virtual circumference. It is a flowchart which shows the prefix bit length determination operation | movement of the non-bootstrap node which comprises the distributed management network system which concerns on one embodiment of this invention.

  Hereinafter, a distributed management network system according to an embodiment of the present invention will be described with reference to the drawings.

  As shown in FIG. 1, a distributed management network system 1 according to an embodiment of the present invention includes a registration server 2, routers 4a and 4b connected to the registration server 2 via a private network 3, and a router 4a. L2 switch 5a connected, L2 switches 5b, 5c connected to router 4b, nodes 6a, 7a-7c connected to L2 switch 5a, nodes 6b, nodes 7d, 7e connected to L2 switch 5b And nodes 7f to 7h connected to the L2 switch 5c.

  The configuration of the distributed management network system 1 shown in FIG. 1 is merely an example, and the distributed management network system of the present invention only needs to include a plurality of nodes including at least one bootstrap node.

  Each router 4a, 4b is comprised by the general router for comprising a subnetwork. Moreover, each L2 switch 5a-5c is comprised by the general L2 switch.

  In FIG. 1, two routers 4 a and 4 b and three L2 switches 5 a to 5 c are illustrated, but the router and the L2 switch are appropriately provided depending on the network configuration of the distributed management network system 1.

  In the present embodiment, each of the nodes 6a and 6b constitutes a bootstrap node. Hereinafter, the nodes 6a and 6b are collectively referred to as “bootstrap node 6”, and each of the nodes 7a to 7h is other than the bootstrap node. A non-bootstrap node is configured and is hereinafter collectively referred to as “non-bootstrap node 7”, and the bootstrap node 6 and non-bootstrap node 7 are collectively referred to as “node”.

  The registration server 2 is a communication module for communicating with a central processing unit (CPU), a random access memory (RAM), a read only memory (ROM), a hard disk device, and the nodes 6 and 7 via the private network 3. It is comprised by the general computer apparatus provided with these.

  A program for causing the computer device to function as the registration server 2 is stored in the ROM or the hard disk device. That is, the CPU functions as the registration server 2 when the CPU executes a program stored in the ROM or the hard disk device using the RAM as a work area.

  The hard disk device of the registration server 2 stores a bootstrap list 8 in which URIs (Uniform Resource Identifiers) of the bootstrap nodes 6 are listed. The bootstrap list 8 is constituted by a one-dimensional list having the URI of the bootstrap node 6 as an element.

  The bootstrap nodes 6 are basically arranged one by one in one subnet. When the number of nodes in the subnet is small, one bootstrap node 6 may be arranged for a plurality of subnets that are close to each other in the network.

  The sequence when each node constitutes the structured overlay conforms to IETF P2PSIP (see Non-Patent Document 4). That is, when the node participates in the overlay, the registration server 2 is accessed and the bootstrap list 8 is obtained at the time of startup.

  The non-bootstrap node 7 transmits an echo request packet to each bootstrap node 6 based on the bootstrap list 8, and based on TTL (Time to live) or RTT included in the echo response returned in response thereto. Thus, the nearest bootstrap node 6 on the network is selected, and messages and data are transmitted / received via the selected bootstrap node 6.

  The bootstrap nodes 6 corresponding to the elements of the bootstrap list 8 are all in an active state and maintain the state of participating in the overlay for a long time, while the non-bootstrap node 7 is necessary for the overlay. You may join and leave accordingly.

  As shown in FIG. 2, each node includes a CPU 10, a RAM 11, a ROM 12, a hard disk device 13, an input device 14 including a keyboard device and a pointing device, a display device 15, a registration server 2, and other nodes. And a communication module 16 for communicating via the private network 3.

  The ROM 12 and the hard disk device 13 store a program for causing the computer device to function as a node. That is, when the CPU 10 executes a program stored in the ROM 12 or the hard disk device 13 using the RAM 11 as a work area, the computer device functions as a node.

  The hard disk device 13 stores a node ID based on the hash value of the unique attribute information of the node in addition to the data to be distributed and managed in the distributed management network system 1.

  In the present embodiment, each node generates a node ID by overwriting a prefix at the beginning of the hash value of the IP address uniquely and fixedly assigned in the distributed management network system 1, and the generated node ID is stored in the hard disk. It is stored in the device 13.

  If the maximum number MBN (Maximum Bootstrap node Number) of the bootstrap nodes 6 constituting the distributed management network system 1 is 256, the bootstrap node 6 generates an 8-bit prefix. The MBN is a power of 2 that is larger than the number of bootstrap nodes 6 constituting the distributed management network system 1 in consideration of the addition of the bootstrap nodes 6.

  Specifically, as shown in FIG. 3, the bootstrap node 6 determines the bootstrap node number continuously assigned from 1 in the private network 3, that is, the registration number of its own URI in the bootstrap list 8. A prefix is generated by reversing the upper and lower bit positions of a value obtained by subtracting one.

  For example, the bootstrap node 6 overwrites the generated prefix at the beginning of a 160-bit hash value, and stores the hash value with the prefix overwritten in the hard disk device 13 as a node ID.

  As shown in FIG. 4A, each node maps each node according to the node ID on the virtual circumference. 4A shows an example in which only the bootstrap node 6 is mapped in order to facilitate understanding of the invention.

  In FIG. 4A, the bootstrap node Pb (1) registered first in the bootstrap list 8 is mapped to the position corresponding to the node ID overwritten with the prefix of 0x00, and is registered next. The bootstrap node Pb (2) is mapped to a position corresponding to the node ID overwritten with the prefix 0x80, and the bootstrap node Pb (3) registered at the end is a node ID overwritten with the prefix 0x40. It is mapped to the position according to.

  In this example, data from which a hash value greater than the node ID of the bootstrap node Pb (1) and less than the node ID of the bootstrap node Pb (3) is obtained is managed by the bootstrap node Pb (3).

  Similarly, data for which a hash value greater than the node ID of the bootstrap node Pb (3) and less than the node ID of the bootstrap node Pb (2) is obtained is managed by the bootstrap node Pb (2), and the bootstrap node Pb Data for which a hash value greater than the node ID of (2) or less than or equal to the node ID of the bootstrap node Pb (1) is obtained is managed by the bootstrap node Pb (1).

  The non-bootstrap node 7 is configured to generate a variable-length prefix in which some of the leading bits of the prefix of the nearest bootstrap node 6 on the network are the same.

  Specifically, the non-bootstrap node 7 is mapped so that the bootstrap node 6 substantially divides the virtual circumference, that is, the bootstrap node 6 of the distributed management network system 1 When the number NB (Number of Bootstrap nodes) is a power of 2, prefixes having the same bit length are generated. Here, if the index is m, the prefix bit length is m.

  On the other hand, when the bootstrap node 6 is not mapped so as to divide the virtual circumference substantially equally, the non-bootstrap mapped to the region where the length between the bootstrap nodes 6 on the virtual circumference is long. The bit length of the prefix of the node 7 is determined so as to be 1 bit shorter than the bit length of the prefix of the non-bootstrap node 7 mapped in the region where the length between the bootstrap nodes 6 on the virtual circumference is short.

More specifically, the bootstrap node number that is greater than the number of bootstrap nodes 6 mapped so that the bootstrap node 6 substantially divides the virtual circumference (ie, the minimum 2 ^{m that} does not exceed NB). Since the bootstrap node 6 is mapped onto the virtual circumference after the virtual circumference is substantially equally divided, the length with the next bootstrap node 6 is shortened.

Accordingly, the non-bootstrap node 7 has an NCB> 2 ^m (hereinafter also referred to as “first condition”) with respect to the bootstrap node number NCB (Number of Connecting Bootstrap node) of the bootstrap node 6 closest to the network. Is established, a prefix with a bit length of m + 1 is generated.

In addition, the bootstrap node 6 in which the value obtained by adding the bootstrap node number to the minimum 2 ^m that does not exceed NB is equal to or less than NB is the bootstrap mapped on the virtual circumference after the virtual circumference is substantially divided. Since the mapping is performed immediately before the strap node 6, the length from the next bootstrap node 6 is shortened.

Therefore, the non-bootstrap node 7 generates a prefix having a bit length of m + 1 when 2 ^m + NCB ≦ NB (hereinafter also referred to as “second condition”).

  In addition, the non-bootstrap node 7 generates a prefix having a bit length of m when NCB and NB satisfy neither the first condition nor the second condition.

  For example, in FIG. 4B, the bootstrap node Pb (1) and the bootstrap node Pb (2) are mapped so as to substantially bisect the virtual circumference, and the bootstrap node Pb (3 ) Is mapped onto the virtual circumference after the virtual circumference is substantially divided into two equal parts.

  Therefore, the non-bootstrap node Pn (3) with NCB 3 satisfies the first condition, and the bit length is m + 1, that is, the prefix “01” is generated from the first two bits of the bootstrap node Pb (3). To do.

  Further, since the non-bootstrap node Pn (1) with NCB of 1 is mapped immediately before the bootstrap node Pb (3), the second condition is satisfied and the bit length is m + 1, that is, the bootstrap node Pb ( A prefix “00” is generated from the first two bits of 1).

  Further, since the non-bootstrap nodes Pn (2a) to Pn (2c) having NCB of 2 satisfy neither the first condition nor the second condition, the bit length is m, that is, the bootstrap node Pb (2) A prefix “1” is generated from the first bit.

  The prefix bit length determination operation of the non-bootstrap node 7 configured as described above will be described with reference to the flowchart shown in FIG. The non-bootstrap node 7 sets the initial value of the index m to 0 (step S1) and executes the following processing.

First, the non-bootstrap node 7 determines whether or not the number NB of bootstrap nodes 6 is 2 ^m (step S2). If it is determined that NB is 2 ^m , the non-bootstrap node 7 sets the prefix bit length plex_p to m (step S3), and ends the prefix bit length determination operation.

On the other hand, when it is determined that NB is not 2 ^m , the non-bootstrap node 7 determines whether 2 ^m <NB <2 ^{m + 1} is satisfied (step S4). If it is determined that 2 ^m <NB <2 ^{m + 1} is not established, the non-bootstrap node 7 increments m (step S5) and executes step S2.

On the other hand, if it is determined that 2 ^m <NB <2 ^{m + 1} holds, the non-bootstrap node 7 has a bootstrap node number NCB of the nearest bootstrap node 6 on the network larger than 2 ^m , that is, , NCB> 2 ^m , or 2 ^m + NCB ≦ NB is determined (step S6).

If it is determined that either one of the conditions NCB> 2 ^m or 2 ^m + NCB ≦ NB is satisfied, the non-bootstrap node 7 sets the prefix bit length plefix_p to m + 1 (step S7). ), The prefix bit length determination operation is terminated.

On the other hand, if it is determined that neither of the conditions NCB> 2 ^m or 2 ^m + NCB ≦ NB is satisfied, the non-bootstrap node 7 sets the prefix bit length plex_p to m (step S3), and sets the prefix. The bit length determination operation ends.

  As described above, the distributed management network system 1 according to the embodiment of the present invention can perform mapping so that the nodes are not concentrated on the virtual circumference, thereby suppressing the deviation in the amount of data managed by the nodes. Can do.

  In addition, the distributed management network system 1 can be applied to a small-scale network in a three-dimensional environment in order to determine a node for managing data after mapping nodes on a virtual circumference.

  The distributed management network system of the present invention can be used as a system for managing data distributed to a plurality of nodes in a structured overlay network.

DESCRIPTION OF SYMBOLS 1 Distributed management network system 2 Registration server 3 Private network 4a, 4b Router 5a-5c L2 switch 6, 6a, 6b, Pb Bootstrap node 7, 7a-7h, Pn Non-bootstrap node 8 Bootstrap list 10 CPU
11 RAM
12 ROM
13 Hard Disk Device 14 Input Device 15 Display Device 16 Communication Module

Claims (4)

A plurality of nodes including at least one bootstrap node and connected to a network;
The node has a node ID based on a hash value of unique attribute information, is mapped according to the node ID on a virtual circumference, and is mapped on the virtual circumference; and the virtual circumference In the distributed management network system in which a node that manages the data is determined based on the position of the data mapped according to the hash value above,
The bootstrap node overwrites a part of the head of the node ID with a uniquely assigned fixed-length prefix,
Among the nodes, the non-bootstrap nodes other than the bootstrap node are a part of the head of the node ID with a variable length prefix having the same part of the head of the prefix of the bootstrap node closest to the network. Overwrite
The bit length of the prefix of the non-bootstrap node is such that the bootstrap node closest to the non-bootstrap node on the network is mapped on the virtual circumference and the non-bootstrap node on the virtual circumference. A distributed management network system, characterized in that the distance between the bootstrap node and the position of the bootstrap node mapped so as to be sandwiched between the bootstrap node is determined so as to become shorter.   The bootstrap node is a part of the beginning of the node ID with a prefix obtained by inverting the bit position of a value obtained by subtracting 1 from the bootstrap node number continuously assigned from 1 in the network. The distributed management network system according to claim 1, wherein: is overwritten.   The bit lengths of the prefixes of the non-bootstrap nodes are determined to be all equal when the bootstrap nodes are mapped so as to substantially divide the virtual circumference. Item 3. The distributed management network system according to Item 2.   If the bootstrap node is not mapped so as to substantially divide the virtual circumference, a non-bootstrap mapped to a long region between the bootstrap nodes on the virtual circumference The bit length of the prefix of the node is determined so as to be 1 bit shorter than the bit length of the prefix of the non-bootstrap node mapped in the region where the length between the bootstrap nodes is short on the virtual circumference. The distributed management network system according to claim 2 or 3.

Images