JP2009151403A - Storage node, cluster storage system and operation method thereof - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an information network distance optimization technique between a client and a storage storing objects, which is applied to a cluster storage system having a network infrastructure.
A storage system is composed of at least two storage nodes and a network, and the storage node has means for measuring an information network distance between another storage node and a client connected to the network. When a client is connected in this storage system, an object stored in each storage node is extracted by searching using an object query required by the client, and an information network of the storage node is extracted. The extracted objects are aggregated to the storage node determined to be close to the client by the distance measuring means.
[Selection] Figure 5

Description

The present invention relates to a technology for improving access performance in a cluster storage system.

  In recent years, as information is digitized, the amount of digitized information is explosively increasing. Along with this, an increase in information management cost is expected to become a bigger problem in the future. Under these circumstances, storage systems such as Network Attached Storage (NAS) and Storage Area Network (SAN), which centralize information management by storage nodes with network connection functions, are widely used to reduce information management costs. is doing.

  As a method of operating network storage such as NAS and SAN, all data created by the client is moved to another high-speed accessible server as the client moves, optimizing data access between the client and the storage server. (For example, refer to the prior art in Patent Document 1).

JP 2003-140930 A

  Patent Document 1 assumes two points: having a network base that can withstand large-scale data transfer, which is transfer of all data held by the client, and relatively low movement frequency of the client. That is, in order to implement the invention described in Patent Document 1, an economical cost is required for the network infrastructure. In addition, even if the cluster storage system has a network infrastructure that can sufficiently withstand the load, there is a problem that the effect cannot be obtained in application to a consumer cluster storage system in which the movement of clients is relatively large.

  An object of the present invention is to provide a storage node, a cluster storage system, and an operation method thereof that can be applied even when a network with a low data transfer rate is used or when a client moves frequently.

  In order to solve the problems of the prior art, the present invention distributes necessary objects in n (n is an integer of 2 or more) storage nodes and arranges m objects (m Is an integer that satisfies 1 or more and satisfies n ≧ m). Specifically, the following cluster storage system (1) and operation method (2) are used.

(1) Cluster storage system.
A cluster storage system consisting of a management host and n storage nodes. The management host has association means for holding association information between the metadata characterizing the object and the object and searching for the object based on the metadata, and the storage node accepts information from the client and information storage means for storing the object Input / output means for inputting the search condition to the association means and outputting or receiving the object to / from the client, and information network distance measuring means for measuring the information network distance between other nodes.

(2) A cluster storage operation method that aggregates objects closer to the client.
When the storage node receives a search condition from a client, the storage node searches for an object in the own storage node that matches the search condition using the association unit, and the search result and information managed by the distance information management unit Cluster storage operation method for transferring objects stored in its own storage node to other storage nodes based on the above.

  By operating the cluster storage system shown in (1) with the method shown in (2), objects can be aggregated to a storage node that is closer in terms of information network distance from the client, even on a cheaper network infrastructure than the conventional method. The performance of the cluster storage system can be improved.

  According to the present invention, even when the network base of the cluster storage system to be applied is inexpensive or the client frequently moves, the objects required by the client are aggregated in advance to a storage node that can be accessed at high speed from the client. To achieve high-speed object access. In particular, the present invention is particularly suitable for application to a cluster storage system using a wireless network infrastructure in which the network communication speed is relatively variable and the reliability is relatively low.

  A configuration and an operation example of the cluster storage system according to the present invention will be described in detail below. In the present specification, all the components assigned the same reference numbers are the same components.

FIG. 1 shows an example of the configuration of a cluster storage system 100 according to the present invention.
In FIG. 1, in the cluster storage system 100, 101 is a client, 103 is a storage node, 105 is a network, and 107 is an object. The client 101 is detachable from the network 105, and the client 101 is disconnected from the network 105 in FIG. In addition, it is assumed that all the storage nodes 103 connected to the network 105 permit the client 101 to access the object 107. This permission method may use whitelist authentication by registering the device identifier of the client 101 or the storage node 103 in the database engine 353 (FIG. 3) of the storage node 103, A configuration may be adopted in which all connected devices can be freely used. A horizontal line of the network 105 indicates a network base, and a vertical line indicates a path from the client 101 and the storage node 103 to the network base. The length of the route indicates the information network distance of the route in the client 101.

Here, the information network distance will be described. The information network distance d is defined as d = l + m / b using the latency l, the transmission speed b of the transmission path, and the average object size m as a scale indicating the speed of object acquisition. Strictly speaking, the latency l and the transmission rate b are not independent, but here, it is assumed that the influence of the transmission rate b on the latency l is sufficiently small. The average used object size m is a scale representing the object size necessary for the client to use the object. The average used object size is a value determined by the property and object size of an object generally used by the client, and is a value determined by the property of the client. The smaller the information network distance d, the shorter the time from when an access request is made to the storage node until the client can actually start processing. That is, in the operation of the storage system, it is desirable that the information network distance d between the client and the storage storing the object is small. There are two types of objects: block objects and stream objects. A block object is a text file or program that can be used by the client holding it from the beginning to the end of the object. On the other hand, the stream type object is an audio or video file. A stream type object is an object that can be used by the client holding only a part of the object necessary for encoding. Due to the nature of these objects, the block object average object size is generally large, while the stream object average object size is sufficiently small. The information network distance is governed by the transmission speed b when the average object size is large, and conversely by the latency l when it is sufficiently small. In other words, the information network distance can be said to be a scale representing the ease of obtaining a comprehensive object that absorbs the difference in the properties of the objects used by the client. The information network distance that is important for object acquisition varies depending on factors that cannot be visually observed, such as path failures, storage node loads, and processing capabilities. For this reason, the ease of acquiring an object cannot be evaluated simply by being close or far in a simple physical linear distance.
Hereinafter, when the network configuration is illustrated in this specification, the notation based on the information network distance is used, and the distance measure such as near / far is also evaluated based on the information network distance. In actual operation, if the storage to store the objects accessed by each client based on this information network distance is not operated as short as possible, the apparent client processing will be slowed down, or this apparent processing time will be shortened. As a result, it is necessary to increase the data processing capacity of the client, resulting in an economic loss.

  The client 101 is a device that newly creates, reproduces, or edits the object 107. Specifically, the client 101 is a personal computer, a mobile phone with a music reproduction function, a digital camera, or the like. The storage node 103 is an apparatus that stores the object 107 and metadata created by the client 101. This metadata is data representing the characteristics of the object 107. For example, the object name and creation time, and the client 101 having a high-performance metadata management function such as GPS extracts location information as metadata.

  FIG. 2 shows an example of the configuration of the client 101. The client 101 includes a user interface 201, an MPU 203, a memory 205, and a network interface 207. The user interface 201 includes an output device 211 such as a display and an input device 213 such as a keyboard. The user interface 201 is an interface for outputting the object 107 to the user or inputting a command by the user. The MPU 203 is a device for executing and processing a program stored in the memory 205. In the client 101, the MPU 203 executes the object processing program 251 to reproduce and edit the object 107 such as image and moving image data, and executes the metadata management program 253 to extract and manage the metadata of the object 107. . The network interface 207 is an interface for exchanging the object 107 and its metadata with the storage node 103 connected to the network 105. An operation example of the client 101 will be described. First, the client 101 acquires the object 107 from the storage node 103 using the network interface 207 or creates a new object 107 using the user interface 201 and the object processing program 251. Next, the acquired object 107 is reproduced or edited using the object processing program 253 and the user interface 201. Thereafter, metadata is assigned using the metadata management program 253, the object 107 and its metadata are transmitted to the external storage node 103 connected to the network using the network interface 207, and the operation of the client 101 ends.

  FIG. 3 shows an example of the configuration of the storage node 103. In this embodiment, a management host (not shown) is not provided, and the configuration example is shown on the assumption that the object 107 and its metadata are managed by the storage node 103. The storage node 103 includes an MPU 203, a memory 205, a network interface 207, and a storage 301. The MPU 203 uses the information network distance measurement program 351 stored in the memory 205 to measure the information network distance between the path between the client 101 and the other storage node 103, and uses the file system 352 stored in the memory 205. The object 107 stored in the object storage unit 311 in the storage 301 is managed. Further, the management of metadata stored in the metadata storage unit 313 in the storage 301 using the database engine 353 stored in the memory 205 and the ranking of numerical data such as information network distance using the numerical data ranking program 354 To do. An operation example of the storage node 103 will be described. The operation of the storage node 103 varies greatly depending on whether the request from the client 101 is a storage request or a reception request. In the case of a storage request, the object 107 and its metadata are transmitted from the client 101 to the storage node 103. The object 107 received using the file system 352 is stored in the object storage unit 311, the metadata received using the database engine 353 is stored in the metadata storage unit 313, and the series of operations ends. On the other hand, in the case of a reception request, a query and output destination information are transmitted from the client 107 to the storage node 103. This query is metadata for specifying the object 107. The storage node 103 inputs the received query to the database engine 353 to search for the necessary object 107 and accompanying metadata. Thereafter, the object 107 and accompanying metadata are transmitted to the client 101 or other storage node 103 specified by the output destination information using the network interface 207.

  In modes other than the present embodiment, an information network distance measurement / management function using the information network distance measurement program 351 and an access function to the object 107 using the file system 352 may be provided to the client 101 or the management host. In a mode in which these functions other than the storage node 103 hold these functions, the client 101 sends a query to the management host, and the client 101 directly accesses the object 107 stored in the storage node 103 by the file system 352 of the client 101. Take measures such as. However, even if the object 107 acquisition means for the storage node 103 is different, the client 101 still acquires the object 107 from the storage node 103 via the network 105. Therefore, it is desirable that the information network distance from the client 101 to the storage node 103 holding the object 107 is short and the object can be accessed at high speed.

  Hereinafter, a method of moving the target object 107 in advance from the client 101 to the storage node 103 having a short information network distance realized in the present invention will be described.

  The feature of the present invention is that when the client 101 is connected, the storage node 103 is judged whether it is close to or far from the client 101 in terms of the information network distance, and the utilization rate of the storage node 103 close to the client 101 is improved. It is to improve the performance. Hereinafter, the storage node 103 close to the client 101 is referred to as the near node 103a, and the other nodes are referred to as the far node 103b. FIG. 4 shows a state in which the client 101 is connected to the network 105 and participates in the cluster storage system 100 from the state of FIG. When the client 101 joins the cluster storage system 100, the near node 103a and the far node 103b are selected by the following procedure. Here, the near node 103a is defined as m (m is an integer satisfying 1 ≧ n and satisfying n ≧ m) of n (n is an integer of 2 or more) storage nodes 103 and far nodes. 103b is defined as (nm) storage nodes 103 other than the near node 103a in the cluster storage system 100. In this embodiment, the operation will be described assuming that n = 5 and m = 2.

  When the client 101 is connected to the network 105, the storage node 103 in the range 401 close to the client 101 becomes the near node 103a, and the other storage nodes 103 become the far node 103b. Thereafter, each far node 103b identifies the near node 103a closest to itself. The method for selecting the near node 103a will be described in detail. When the client 101 is connected to the network 105, the client transmits a network connection signal 403 to all the storage nodes 103. This connection signal 403 includes the identifier of the client 101 and the address of the client 101 on the network 105. Upon receiving the connection signal 403, the storage node 103 measures the information network distance with the client 101 using the information network distance measurement program 351, and notifies the other storage nodes 103 of the measurement result. Each storage node 103 compares the measured information network distance with the information network distance of the other storage node 103 and the client 101 using the information network distance measurement program 351, and determines whether its own information network distance is in the upper m rank. ) The storage node 103 that has been determined to be the upper m rank becomes the near node 103a, and the other storage nodes 103 become the far nodes 103b. The first-ranked near node 103a, that is, the storage node 103 closest to the client 101 notifies the client 101 of information necessary for transmission / reception of the object 107 thereof. In the embodiment in this specification, the object 107 newly created by the client 101 is temporarily or permanently stored in the near node closest to the client 101. Therefore, it is necessary for the client 101 to grasp the nearest near node 103a by this operation. After selecting the near node 103a, the far node 103b transmits a distance measurement request to the surrounding storage nodes 103 in order to measure the distance to the other storage nodes 103. At this time, the near node 103a returns to the transmission source in response to the distance measurement request, but the far node 103b does not respond even if it receives the distance measurement request. By this procedure, the far node 103b collects information network distances of all the near nodes 103a, and specifies the nearest near node 103a among them. By this operation, the transfer is performed with a combination that minimizes the information network distance between the far node 103b and the near node 103a, so that the transfer of the object 107 becomes faster.

The above is the common operation of the cluster storage system 100 according to the present invention.
Through the above operation, the storage node 103 that is relatively close to the client 101 is selected. If there is a target object in the storage node 103 within the above range, the client 101 can access the object 107 at high speed. Hereinafter, an operation of consolidating the objects 107 in this close range will be described. Since the object 107 aggregation operation differs depending on the mode of implementation, each will be described.

An aspect in which the object 107 aggregation operation is an object copy from the far node 103b to the near node 103a will be described.
FIG. 5 is a diagram for explaining an operation of aggregating the objects 107 by copying to the near node 103a from the state in which the client 101 is connected to the cluster storage system 100 shown in FIG. The client 101 notifies the query 501 of the required object 107 to each storage node 103 that has recognized whether it is the near node 103a or the far node 103b. Each storage node 103 that has received the query 501 from the client 101 searches the object 107 using the database engine 353. Further, each fur node 103b copies the object 107 corresponding to the query 501 and transmits it to the near node 103a specified by each fur node 103b by the method of the first embodiment (503). Through the above procedure, the objects 107 required by the client 101 are collected in the near node 103a located in a range close to the client 101. The client 101 can acquire the object 107 at high speed by accessing the object 107 of the near node 103a. The object 107 newly created by the client 101 or moved from the storage device in the client 101 to the storage system 100 is stored in the near node 103a closest to the client 101. In the cluster storage system 100 according to the present invention, when the client 101 connects to the cluster storage system 100 from various positions, since the nearest near nodes 103a are selected in a distributed manner, a distributed arrangement state of the objects 107 can be realized.

Next, the operation when the client 101 disconnects 601 from the network 105 from the state where the objects 107 described in FIG. 5 are aggregated in the near node will be described with reference to FIG. When the client 101 is disconnected from the network 105, the near node 103a that aggregates the objects 107 performs any of the following operations on its own object 107.
The newly created object 107a is stored in itself.
The updated object 107b is returned to the original far node 103b and updated.
The object 107c that has not been updated is discarded.

  By the object operation, the consistency of the object 107 in the entire cluster storage system 100 is ensured by the minimum movement of the object 107. The operation when the client 101 is disconnected will be described in detail. First, when disconnecting from the network 105, the client 101 notifies all storage nodes 103 of a network disconnection signal. As a method of detecting that the client 101 is disconnected from the network 105, there is also a method of generating a connection signal periodically and determining that the client 101 is disconnected from the network 105 when the connection signal cannot be confirmed for a certain period. The near node 103a that has detected the network disconnection of the client 101 first determines whether there is a newly created object 107a. If there is a newly created object 107a, it is stored in its own storage 301 as it is. Next, in the object aggregation operation at the time of client connection, it is determined whether there is an object 107 received from another far node 103b. The received object 107 has information for specifying the far node 103b that is the transmission source. If there is a received object 107, it is further determined whether the object 107 is an updated object 107b. The update is confirmed by a method of comparing the reception time of the object 107 with the update time. If an update has occurred in the object 107 received from another far node 103b, the same update is required for the original object 107b, so the object 107b is returned to the original far node 103b 603, and the far node 103b The corresponding object 107 is updated. On the other hand, when the object 107 has not been updated, the original object 107 does not need to be updated, and is discarded without reply. As described above, when the processing of the object 107 after disconnecting the client 101 is completed, the near node 103a and the far node 103b are not distinguished, and the session with the client 101 is ended.

  In the present embodiment, an aspect will be described in which a balance function based on a storage capacity load is stored when an object 107a newly created by the client 101 is stored in the storage node 103.

  The operation when the client 101 disconnects from the network 105 from the state in which the objects 107 in FIG. 5 are aggregated in the near node 103a will be described with reference to FIG. When the client 101 is disconnected from the network 105, all the storage nodes 103 notify the other storage nodes 103 of their capacity load. Each storage node 103 summarizes the capacity load of all storage nodes 103 connected to the cluster storage system 100 in a table 701. In this embodiment, node IDs are assigned in order from the left of the storage node 103 in FIG. 7 for convenience. The node ID may be a unique identifier given when the storage node 103 is manufactured, or may be an identifier given when connected to the network 105. This is expressed by being added to the storage node 103 in the figure. In FIG. 7, the storage node 103 with the lowest capacity load is the leftmost node ID 103. In the second embodiment, when the client 101 is disconnected, the newly created object 107a is stored in the near node 103aa closest to the client 101. However, in this embodiment, the newly created object is transmitted to the storage node 1 with the smallest capacity load. The storage node No. 1 is notified to the client 101, and the storage node No. 1 receives and stores the newly created object. Depending on the operation of the cluster storage system 100, the connection position of the client 101 may not change. At this time, in the new object 107a storing method described in the second embodiment, the new object 107a is always stored in one storage node 103, and a capacity load is applied to one storage node 103 in a concentrated manner. When the storage node 103a with such a large capacity load is determined as the near node 103a by the connection of another client 100, the object aggregation operation may not be realized. The storage operation of the new object 107a described in the present embodiment can reduce the local and permanent capacity load in the cluster storage system 100 according to the present invention, and can improve the utilization efficiency of the cluster storage system 100. On the other hand, for objects 107 other than the new object 107a aggregated when the client 100 is connected, it is determined whether there is an object 107 received from another far node 103b in the object aggregation operation when the client 101 is connected as in the second embodiment. . If there is a received object 107, it is further determined whether the object 107 is an updated object 107b. The update is confirmed by a method of comparing the reception time of the object 107 with the update time. If an update has occurred in the object 107 received from another far node 103b, the same update is required for the original object 107b, so the object 107b is returned to the original far node 103b 603, and the far node 103b The corresponding object 107 is updated. On the other hand, when the object 107 has not been updated, the original object 107 does not need to be updated, and is discarded without reply. As described above, when the processing of the object 107 after disconnecting the client 101 is completed, the near node 103a and the far node 103b are not distinguished, and the session with the client 101 is ended.

In this embodiment, a mode of exchanging objects at the time of object aggregation will be described. When the object aggregation operation described in the second and third embodiments occurs, the capacity load of the near node 103a becomes higher than that of the far node 103b. For this reason, if the arrangement of the objects 107 is biased or the capacity load of the entire cluster storage system 100 is high, the objects 107 may not be aggregated in the near node 103a. In the present embodiment, an unnecessary increase in the capacity of the near node 103a can be suppressed by returning an unnecessary object 107 from the near node 103a to the far node 103b. Moreover, it is desirable that the replacement of the object 107 is simultaneous replacement. The simultaneous exchange referred to here is an exchange technique in which the object 107 is transmitted only for the received amount in a state in which the connection between the transmission side and the reception side is not performed as in the peer-to-peer connection. By performing simultaneous exchange, an instantaneous increase in capacity load can be suppressed, so that even when the object 107 to be exchanged is huge, an object aggregation operation can be executed while suppressing an increase in capacity load. Hereinafter, an object aggregation operation for exchanging objects will be described in detail.

  FIG. 8 is a diagram for explaining the object aggregation operation from the state in which the client 101 in FIG. 4 is connected to the cluster storage system 100. The client 101 notifies each storage node 103 of the query 501 of the object 107 that is required. Each storage node 103 generates its own object suitability list (not shown) from the query 501 notified from the client 101 and notifies this to other storage nodes 103. The degree of matching between the object 107 and the query 501 is a value calculated by h / q when the number of metadata matching the object 107 out of q metadata included in the query 501 is, for example. All the storage nodes 103 hold the matching list 801 of the objects 107 in the entire cluster storage system 100 by combining the object matching lists sent from the other storage nodes 103. When the far node 103b holds the object 107d within the upper r rank, the far node 103b copies the object 107d and transmits it to the nearest near node 103a. The near node 103a stores the received object 107d having a high relevance rate in itself, and the near node 103a receiving the object 107d having a high relevance rate with the object 107d has a relevance smaller than the top r among the objects 107 possessed by the near node 103a. The object 107e with a low precision is returned in exchange. The amount of objects with a low relevance rate returned at this time is determined based on the amount of objects with a high relevance level received, the capacity load of the near node 103a itself, and the like. The above is the object aggregation operation when the client 101 is connected in the fourth embodiment.

  Further, an operation when the client 101 is disconnected from the network 105 will be described. FIG. 9 is a diagram for explaining the operation when the client 101 disconnects from the network 105 from the state in which the objects 107 of FIG. When the storage node 103 detects that the client 101 has been disconnected from the network 105, the far node 103 b that has received the object 107 from the other storage node 103 returns the received object 107 to the original storage node 103. By this object 107 return operation, the state in which the objects 107 are distributed is held. The newly created object 107a in the near node 103a may be stored in the near node 103aa closest to the client 101 by the method described in the second embodiment, or the capacity load of the storage node 103 of the cluster storage system 100 by the method described in the third embodiment. May be stored in a small storage node 103. Further, the object 107 aggregated from the far node 103b returns the updated object 107b to the original storage node by the method described in the second and third embodiments, and the object 107c that has not been updated is discarded.

  In this embodiment, a mode in which the mesh network 1005 is used for the network 105 will be described. FIG. 10 shows a cluster storage system 1000 to which a mesh network 1005 is applied. Such a mesh network 1005 is, for example, a wireless network. A wireless network can be connected anywhere within the reach of radio waves, and the quality of communication is likely to change due to the influence of distance and obstacles. That is, it can be said that the connection position of the client 101 is easy to change and the information network distance between the client 101 and the storage node 103 is easy to change. In the mesh network 1005, a storage node 103c that cannot directly communicate with the client 101 may be connected to the storage system 1000 via another storage node 103.

  Hereinafter, the storage node 103 that cannot directly communicate with the client 101 is referred to as an “alone node 103c”. An object aggregation operation in the network 1005 in which the alone node 103c exists will be described.

First, when the client 101 is connected to the storage system 1000, the near node 103a and the far node 103b are specified for the storage node 103 that can directly communicate with the client 101 by the near node specifying operation in the present invention. However, at this point, the alone node 103c does not recognize whether it is the near node 103a or the far node 103b.
Due to the feature that direct communication with the client 101 is not possible, the alone node 103c is always the far node 103b. That is, in order to obtain the effect by applying the present invention in the storage system 1000, it is only necessary to recognize that the alone node 103c is the far node 103b, identify the near node 103ab closest to itself, and aggregate the objects 107.

  Hereinafter, a method of recognizing that the alone node 103c is the far node 103b, specifying the near node 103ab closest to itself, and aggregating the objects 107 will be described.

  Each storage node 103 that has recognized that it is the far node 103b by the method described in the first embodiment sends an information network distance measurement request to all the storage nodes 103 connected to the network 1005 in order to detect the near node 103a closest to itself. Send. In the first embodiment, only the near node 103a responds to this request, but in this embodiment, the storage node 103 that has not received the connection signal of the client 101, that is, the alone node 103c also responds thereto. By this operation, the far node 103b recognizes that the alone node 103c exists, and the alone node 103c recognizes that the own node 103c is a kind of far node.

  The alone node 103c recognizes itself as the far node 103b, and transmits an information network distance measurement request to all the storage nodes 103 connected to the network 1005. Then, the near node 103a capable of directly communicating with the alone node 103c responds to this request, and the alone node 103c identifies the near node 103a closest to itself.

  By this operation, the alone node 103c recognizes that it is the far node 103b, and specifies the near node 103a closest to itself by the same method as the far node 103b.

  Further, the alone node 103c notifies the nearest near node 103a that it is an alone node, and the near node 103ab that aggregates objects from the alone node 103c receives a query 501 received from the client 101 during the object aggregation operation. Send to mediate to. The alone node 103c that has received the query 501 transmits the object 107a having a high fitness to the query 501 among the objects 107 that it has, to the nearest near node 103a identified by the above operation. As a result, the objects 107a having a high degree of conformity to the query required by the client 101 are collected in the near node 103a, and the object access in the cluster storage system 1000 is accelerated.

  In the present embodiment, a mode in which the present invention is applied to a mesh network 1005 in which there is no near node 103a that can directly communicate with the alone node 103c will be described. FIG. 11 shows a cluster storage system 1000 to which a mesh network 1005 in which the near node 103a and the alone node 103c cannot communicate directly is applied. Unlike the case of FIG. 10, in the storage system 1000, there is no near node 103a that can communicate with the alone node 103c, which means that there is no storage node 103 that responds to the near node 103a selection operation by the alone node 103c described in the fifth embodiment. ing. In such a situation, the far node 103bb that has detected the alone node 103c operates as a mediation node. The intermediary node is a far node 103b that transmits the request of the client 101 to the alone node 103c and transmits the object 107 from the alone node 103c to the selected near node 103a.

  Hereinafter, the near node 103a specifying operation in which the far node 103b becomes the mediation node 103bb in order to access the object 107 stored in the alone node 103c at high speed will be described in detail. The alone node 103c recognizes itself as the alone node 103c by the specific operation of the alone node 103c described in the fifth embodiment. Then, as in the fifth embodiment, the alone node 103c transmits an information network distance measurement request in order to identify the near node 103a that aggregates the objects 107. Since only the near node 103a responds to this information network distance measurement request, when there is no near node 103a connected to the alone node 103c, there is no response to the alone node 103c. When there is no response for a certain time, the alone node 103c measures the 103b information network distance to the far node 103b that transmitted the information network distance measurement request in the near node 103a specifying operation of the far node 103b. As a result of the measurement, the far node 103bb closest to the alone node 103c is specified, and the alone node 103c transmits a mediation request signal to 103bb. Then, the far node 103bb that has received the mediation request signal becomes the mediation node 103bb. Then, the alone node 103c sends an object having a high conformity to the query to the mediation node 103bb to the mediation node 103bb. Send.

  This mediation node allows the client 101 to access the object 107 that exists in the storage node 103c that cannot directly communicate with the client 101 and the near node 103a.

  Thereafter, the object aggregation operation can be realized by applying the first to third embodiments.

It is an example of the initial state of the storage system by this invention. It is a structural example of the client in this invention. It is a structural example of the storage node in this invention. It is an example of a network configuration when a client connects to a network. 12 is an example of an object aggregation operation in the second embodiment. 10 is an operation example when the client is disconnected in the second embodiment. 10 is an operation example when the client is disconnected in the third embodiment. 10 is an example of object aggregation operation in the fourth embodiment. 12 is an operation example when the client is disconnected in the fourth embodiment. 10 is an example of an operation of recognizing an alone node in the fifth embodiment. 12 is an example of an operation for recognizing an intermediary node in the sixth embodiment.

Explanation of symbols

100,1000: Cluster storage system
101: Client
103: Storage node
103a: Near node
103b: Farnode
103c: Alone node
105: Network
107: Object
1005: Mesh network

Claims (17)

In a method of operating a cluster storage system comprising a network to which a client can be connected and a plurality of storage nodes connected to the network,
When the plurality of storage nodes detect the connection of the client to the storage system, the information network distance program is measured using the information network distance program,
The plurality of storage nodes send and receive the measurement results to each other, register a near node to a storage node close to the client according to a predetermined condition, and register a far node to the remaining storage nodes,
The far node uses an information network distance program to measure the information network distance with other storage nodes except the own storage node, and stores the near node closest to each own storage node in itself.
The client notifies the storage node of object feature metadata;
The far node searches for an object in the own storage node using its own database engine based on the feature metadata of the object notified from the client, and transmits the searched object to the storage node of the stored near node. A cluster storage operation method characterized by the above. In claim 1,
The plurality of storage nodes compare the transmitted and received measurement results with each other, and register that they are near nodes when the information network distance between the client and the client is within a predetermined number from the smallest. How to operate a cluster storage system. In claim 1,
The data transmitted to the near node has information indicating the far node where the data is stored,
When the near node that has received the data detects that the client has been disconnected from the network, the near node transmits the data to the far node indicated by information included in the data. . In claim 1,
When the plurality of storage nodes are the storage nodes closest to the client from the measurement results transmitted and received from each other, the plurality of storage nodes notify the client,
When the client receives the notification and creates new data, the client stores the new data in a cluster node that has transmitted the notification. In claim 1,
The plurality of cluster storages send and receive capacity load information to each other, create a capacity load table,
The storage node with the smallest capacity load listed in the capacity load table notifies the client,
When the client receives the notification and creates new data, the client stores the new data in a cluster node that has transmitted the notification. In claim 1,
The near node searches for an object in its own storage node using its own database engine based on the feature metadata of the object notified from the client,
An operation method of a cluster storage system, wherein when the object is received from the far node, an object other than the searched object is transmitted to the far node. In claim 1,
The plurality of storage nodes register themselves as being independent nodes when the information network distance measurement by the information program is received from another storage node when the connection of the client to the storage system is not detected. ,
The alone node uses the information network distance program to measure the information network distance with other storage nodes other than the own storage node, and stores the near node closest to the own storage node in itself.
The alone node searches for an object in the own storage node using its own database engine based on the feature metadata of the object notified from the client, and transmits the searched object to the storage node of the stored near node. A cluster storage operation method characterized by the above. In claim 7,
If the near node is not found when the information node distance is measured with another storage node other than its own storage node using the information network distance program, the alone node mediates to the far node. Send a request signal,
The far node that has received the mediation request signal transfers the feature metadata to the alone node,
The alone node that has received the feature metadata searches for an object in its own storage node, and transmits the searched object to the storage node of the far node that is the transmission destination of the mediation request signal.
The far node that has received the object transmits the object received from the alone node to the storage node of the stored near node. Storage means for storing objects;
Search means for searching for an object stored in the storage based on feature metadata;
A network interface for network connection;
Client detection means for detecting whether the client is connected to the network;
A storage node comprising an information network distance measuring means for measuring an information network distance with a client or other storage node on the network,
Upon detecting that a client is connected to the network to which the storage node is connected, measure the distance to the client,
Send / receive the measured result to / from other storage nodes connected to the network, register itself as a near node or a far node according to a predetermined condition,
If registered as the far node, measure the information network distance with other storage nodes, select and store the near node based on the measurement results,
Search for an object in the storage means of itself based on the feature metadata received from the client,
A storage node that transmits the searched object to the stored near node. In claim 9,
Among the storage nodes that have transmitted and received the measurement results, when the information network distance from the client is within a predetermined rank, register itself as a near node,
A storage node characterized by registering itself as a far node when the predetermined order is exceeded. In claim 9,
When registering as the near node, the object transmitted from the far node is received, stored in the storage means,
When detecting that the client has left the network, the storage node returns or discards the object to the far node based on whether or not the object being stored is updated. In claim 10,
The storage node that notifies the client to transmit a new object to the own storage node when the information network distance from the client is the shortest among the storage nodes that have transmitted and received the measurement result. In claim 10,
When registering as a near node, send and receive capacity load with other near nodes to create a capacity load table,
A storage node characterized by notifying the client to send a new object to its own storage node when the capacity load is the smallest near node. In claim 9,
When registered as the far node,
Search for an object in the storage means of itself based on the feature metadata received from the client,
A storage node, wherein when an object is received from a far node, the object is selected based on the search, and the selected object is transmitted to a far node that is a transmission source of the received object. In claim 9,
Without detecting the connection of the client to the network, if the far node detects a signal for measuring the information network distance,
Register itself as an Alone node,
Measure the information network distance with other storage nodes, select and store near nodes based on the measurement results,
Search for an object in its own storage means based on the received feature metadata,
A storage node that transmits the searched object to the stored near node. In claim 15,
Measure the information network distance with other storage, and when the near node to be measured cannot be detected,
Search for an object in its own storage means based on the received feature metadata,
A storage node, wherein a mediation request signal for requesting the far node to transfer the retrieved object and the object to the near node is transmitted to the far node. A network to which a client can connect; and a plurality of storage nodes connected to the network;
The plurality of storage nodes are:
Storage means for storing objects;
Search means for searching for an object stored in the storage based on feature metadata;
A network interface for network connection;
Client detection means for detecting whether the client is connected to the network;
In a cluster storage system having information network distance measuring means for measuring an information network distance with a client or other storage node on the network,
When the connection of the client to the network is detected, the plurality of storage nodes measure an information network distance with the client using an information network distance program,
The plurality of storage nodes send and receive the measurement results to each other, register a near node to a storage node close to the client according to a predetermined condition, and register a far node to the remaining storage nodes,
The far node uses an information network distance measurement program to measure the information network distance with other storage nodes, and stores the near node closest to each of its own storage nodes in itself.
The storage node receives object feature metadata;
The far node searches for an object in the own storage node using its own database engine based on the feature metadata of the object, and transmits the searched object to the storage node of the stored near node. Cluster storage system to be used.

Images