JP5655044B2 - Data cache system - Google Patents

Description

  The present invention belongs to a so-called general technical field of a cache such as an L1 (level 1) / L2 (level 2) cache in a CPU (Central Processing Unit), a cache installed hierarchically between CPUs and memories, and the like. Since the cache mechanism executes transmission / reception of information via NW (Network), it belongs to the field of cache technology such as HTTP (HyperText Transfer Protocol) data on the Web. Furthermore, since it is assumed that the cache mechanism has a cluster configuration, it also belongs to the field of distributed processing technologies such as distributed databases and MapReduce.

  Conventionally, read / write cache technology has existed for L1 / L2 caches in CPUs. However, since multi-CPUs and multi-cores have become widespread in recent years, there are multiple cache entities. Various algorithms based on the MESI protocol (see Non-Patent Document 1) have been proposed.

  However, the cache mechanism on the CPU side assumes update detection based on bus snoop monitoring and cannot be applied as it is in an NW environment. It is assumed that update notifications are collected and monitored at some specific server or the like, but there is a problem that it is not realistic because it easily becomes a bottleneck.

  For this reason, the cache mechanism in the NW environment uses the information on the client side as it is within the lifetime based on the TTL (Time To Live) value, such as DNS (Domain Name System). Can be reduced at the same time as the response time can be greatly shortened, or when there is no change in the data as in the case of HTML (HyperText Markup Language) data acquisition, Not Modified is received and the client side A method that can reduce the load on the server side by using the existing data as it is has been generally used.

  However, in the cache mechanism using the TTL value, not updating the information held on the client side until the lifetime of the TTL value imposes a limit on the trigger of the update reflection, while in the middle of the lifetime of the TTL value. When updating, there was a problem that the maintenance of consistency had to be sacrificed. Regarding the Not Modified response, if the data size is certainly large, such as several tens of MB, there is an effect of reducing the server load, but the data size handled as in communication processing is small, such as several kilobytes. In this case, the effect is small and the response itself is returned via the NW, so that there is a problem that the response time can hardly be shortened.

  On the other hand, in the technology of JBossCache (see Non-Patent Document 2), a dedicated cache instance is held in a client application, and when the data is updated via the cache instance in the application, this cache instance is transferred to another client. It is possible to realize behavior close to cache coherence in the CPU by automatically performing update notification to the cache instance.

T.Suh, D.M.Blough, et al., “Supporting cache coherence in heterogeneous multiprocessor systems”, Design, Automation and Test in Europe Conference and Exhibition 2004, 1530-1591 / 04, IEEE “JBossCache”, [online], [searched August 24, 2012], Internet <URL: http://www.jboss.org/jbosscache/>

However, in mechanisms such as JBossCache, data is permanently assigned to clients. On the other hand, in the present invention, the client side has a cluster configuration composed of a plurality of servers, and assumes a method in which the data allocation destination is changed due to a server failure or expansion. There are problems (1) and (2).
(1) Since the cluster itself often conceals the internal server by NAT (Network Address Translation) or the like, it is difficult to send an update notification from the outside to the internal server.
In this regard, it is assumed that a session (hole) stretched when connecting to acquire cache data is continued and the session (hole) is diverted. However, this method cannot cope with a case where the client side cluster configuration changes due to a server failure or expansion, and the session is disconnected. Also, in the case of session disconnection when the cluster to which the cache data is called changes in the same manner, the hole for update notification disappears and cannot be handled.
(2) If the server configuration changes due to a failure or expansion within the cluster itself, the data to be referenced or updated will be moved to another server and referenced to the correct destination of the destination. It becomes difficult to send requests and update notifications.

  The present invention has been made in view of such a background, and the present invention updates cache data information while concealing the cluster configuration of the caller (client) and callee (original data storage server) from each other. It is an object of the present invention to provide a data cache system that can notify each other.

  In order to solve the above problem, the invention according to claim 1 is a client cluster in which a plurality of servers that store data as cache data constitute a cluster, and a plurality of servers that store the original data of the cache data. Is a data cache system that is connected to an original data cluster constituting a cluster via a network, and is a first server that is a server included in the client cluster, and a second server that is included in the original data cluster. Each of the servers includes (1) a data key composed of an identifier unique to the data and an address of the original data cluster that stores the original data of the data, and a cache that executes read processing and write processing of the data Manager-specific identifier and relevant A cache manager key configured with an address of a cluster indicating the client cluster or the original data cluster to which a cache manager belongs, metadata storing a cache manager having a read authority and a write authority for the data; (2) a storage unit that stores the data key, and cache data management information in which the data key is associated with unique identifiers of a plurality of cache managers set in the servers in each cluster; A distribution processing unit that extracts a data key attached to the data and refers to the cache data management information to determine the cache manager corresponding to the data; and the cache manager; and Server cache manager When the data reference request is received, the data key attached to the reference request is searched for whether or not the data is stored in the storage unit, and the data is not stored. In this case, the address of the original data cluster is extracted from the data key, and an acquisition request with the data key of the data and the cache manager key of the cache manager is added to the first request message to the original data cluster. The distribution processing unit of the second server extracts the data key from the received first request message and stores the cache manager that stores the original data corresponding to the data. Determining with reference to management information, sending the first request message to the determined cache manager, The cache manager of the second server that has received the first request message refers to the metadata of the data, and extracts the cache manager key from the first request message when it has the write authority. Then, after adding the extracted cache manager key to the metadata read authority of the data, the address of the client cluster is extracted from the cache manager key, and the data is transmitted to the cache manager of the first server. In the data cache system, the cache manager of the first server receives the data, stores the data in its storage unit, and outputs the data.

In this way, when the cache manager of the first server included in the client cluster does not store the data that has received the reference request, the original data of the data in the second server included in the original data cluster Can be received and the data can be received from the cache manager having the write authority and stored in its own storage unit.
At this time, the first server included in the client cluster only needs to transmit the first request message to the address of the original data cluster, and the configuration of the second server in the original data cluster may remain hidden. The distribution processing unit of the second server that has received the first request message determines the cache manager that stores the original data by using the data key, and refers to it by sending the first request message to the determined cache manager Request processing can be performed.

  According to the second aspect of the present invention, when the cache manager of the second server that has received the first request message refers to the metadata of the data and does not have the write authority, the write authority 2. The data cache system according to claim 1, wherein a cache manager having a right is extracted, and the first request message is transmitted to the extracted cache manager having the write authority.

  In this way, when the cache manager that stores the original data does not have write authority, the metadata is referred to, the first request message is transmitted to another cache manager having write authority, and the reference request Processing can be executed.

  According to a third aspect of the present invention, when the cache manager of the first server receives the update request for the data, the update request indicates whether or not the data is stored in its storage unit. When the data key attached to is searched and the data is not stored, or even if the data is stored and the metadata is referred to and the user does not have the write authority, the original data is read from the data key. An address of the cluster is extracted, and an update request with the data key of the data and the cache manager key of the cache manager is included in a second request message to the original data cluster and transmitted, and the second server The distribution processing unit extracts the data key from the received second request message and stores the original data corresponding to the data A cache manager of the second server that has determined the cache manager with reference to the cache data management information, transmits the second request message to the determined cache manager, and has received the second request message. The manager refers to the metadata of the data, updates the data when the manager has the write permission, extracts the cache manager key from the second request message, and sets the read permission of the metadata. After adding the extracted cache manager key, the address of the client cluster is extracted from the cache manager key, the updated data is transmitted to the cache manager of the first server, and the cache of the first server is sent. The manager receives the updated data and stores it Stored in, and the data cache system according to claim 1 or claim 2, wherein the outputting the data.

  In this way, when the cache manager of the first server included in the client cluster does not store the data that has received the update request, the original data of the data of the second server included in the original data cluster Is transmitted and a second request message indicating an update request is transmitted to the cache manager having the write authority to update the data. Then, the cache manager of the first server provided in the client cluster can receive the updated data after obtaining the read authority and store it in its own storage unit.

  According to a fourth aspect of the present invention, when the cache manager of the second server that has received the second request message refers to the metadata of the data and does not have the write authority, the write authority 4. The data cache system according to claim 3, wherein a cache manager having a right is extracted, and the second request message is transmitted to the extracted cache manager having the write authority.

  In this way, when the cache manager that stores the original data does not have the write authority, the metadata is referred to, the second request message is transmitted to another cache manager having the write authority, and the update request Processing can be executed.

  According to the present invention, there is provided a data cache system capable of notifying update of cache data information while concealing the cluster configuration of a caller (client) and a callee (original data storage server) from each other. can do.

It is a figure which shows the structural example of the data cache system which concerns on this embodiment. It is a figure which shows the structural example of the original data cluster which concerns on this embodiment. It is a figure which shows the example in which the server in the client cluster concerning this embodiment processes the reference request of cache data. It is a figure which shows the example in which the server in the client cluster concerning this embodiment processes the update request of cache data. It is a functional block diagram which shows the structural example of the server which concerns on this embodiment. It is a figure which shows the data structure of the cache data management information which concerns on this embodiment. It is a flowchart which shows a process when the cache manager of the server in the client cluster which concerns on this embodiment receives the request | requirement of a data reference request. It is a flowchart which shows a process when the cache manager of the server in the client cluster which concerns on this embodiment receives the request | requirement of a data update request. It is a flowchart which shows a process when the cache manager of the server in the client cluster which concerns on this embodiment receives the request | requirement of a data update request. It is a flowchart which shows the flow of the process in which the cache manager which concerns on this embodiment performs a reception method. It is a figure which shows the example which the cache manager of the server in the client cluster concerning this embodiment acquires data with Read authority. It is a figure which shows the example which the cache manager of the server in the client cluster concerning this embodiment acquires data with a Write authority. It is a figure which shows the example in which the cache manager of the server in the client cluster which concerns on this embodiment has a Write authority, and updates data. It is a figure which shows the example which requests | requires update to the other cache manager with a Write authority, when the cache manager of the server in the client cluster which concerns on this embodiment receives the update request of data. It is a figure which shows the example which transmits an update notification to all the servers in the client cluster which concerns on this embodiment.

  Next, the data cache system 100 in a mode for carrying out the present invention (hereinafter referred to as “the present embodiment”) will be described.

<System configuration>
FIG. 1 is a diagram illustrating a configuration example of a data cache system 100 according to the present embodiment.
As shown in FIG. 1, a data cache system 100 includes an original data cluster 10 configured by a plurality of servers 1 (1y: first server) and storing original data (original data) in a distributed manner, and a client cluster. 20 are connected via a network. The client cluster 20 forms a cluster by a plurality of servers 1 (1x: second server), and acquires response information by transmitting a request (request information) to the original data cluster 10. Each server 1 (1x) of the client cluster 20 stores the information acquired from the original data cluster 10 as cache data.
In the present embodiment, the description will be given on the assumption that the client side is the client cluster 20, but as shown in FIG. 1, the processing can be similarly executed also in the client terminal 30 which is a single server. . Similar to the server 1 (1x) constituting the client cluster 20, the client terminal 30 transmits a request (request information) to the original data cluster 10, and stores the acquired information as cache data.

<Cluster configuration>
Next, configuration examples of the original data cluster 10 and the client cluster 20 will be described. Since both the original data cluster 10 and the client cluster 20 have the same configuration as described below, the original data cluster 10 will be described as an example here.

FIG. 2 is a diagram illustrating a configuration example of the original data cluster 10 according to the present embodiment.
As shown in FIG. 2, the original data cluster 10 includes a load balancer B (Balancer: indicated as “B” in each figure), a plurality of dispatchers D (Dispatcher: indicated as “D” in FIG. 2), a plurality of And a processor P (Processor: written as “P” in FIG. 2).

The load balancer B acquires a request (request information) from the external terminal 40 and transmits response information to the request to the external terminal 40. Further, the load balancer B distributes the acquired request to one of a plurality of dispatchers D (D ₁ , D ₂ , D ₃ ) based on a simple algorithm such as round robin. Here, the external terminal 40 is the server 1 (1x) or the client terminal 30 constituting the client cluster 20 shown in FIG.

Each dispatcher D (D ₁ , D ₂ , D ₃ ) is connected to a plurality of processors P (P ₁ , P ₂ , P ₃ ), and requests received from the load balancer B are sent to the processor P (P ₁ , P ₁ , distributed to any of the _{P 2,} P _3). The dispatcher D analyzes input data, applies a predetermined data management algorithm, for example, a consistent hash method, determines a processor P that is a data storage destination, and transmits the request.

The processor P is connected to a plurality of dispatchers D (D ₁ , D ₂ , D ₃ ), receives a request from one of the dispatchers D, and in accordance with the request, new data is stored in its own storage unit (local storage). Is executed, the existing data is updated, and the data is referenced (acquired). The local storage includes a local cache (cache data storage unit 131 in FIG. 5) that stores data as a cache. In the following description, “local” is used without any particular explanation, and it means a local storage including a local cache. Furthermore, here, it is assumed that each data is stored in the storage unit as an XML (Extensible Markup Language) file.

The dispatchers D (D ₁ , D ₂ , D ₃ ) and the processors P (P ₁ , P ₂ , P ₃ ) are not limited to the three devices illustrated in FIG. . In FIG. 2, the dispatcher D and the processor P are described as separate devices. However, the dispatcher D and the processor P can be operated as separate functions on the same server. In the following description, the two servers 1 (the server 1 on the client cluster 20 side is “server 1x” and the server 1 on the original data cluster 10 side is “server 1y”).

<Overview>
In order to solve the technical problems shown in the above problems (1) and (2), the present invention has the following technical means.
Regarding the problem (1) that the client-side cluster itself conceals the internal server 1 with NAT or the like, it is necessary to enable communication from outside to a machine (server 1 or the like) in the cluster via Push. Here, since the client cluster 20 has a cluster configuration for storing and managing data in the first place, the client cluster 20 has a mechanism that allows a request signal to reach the server 1 in the cluster from the outside. By using the mechanism and describing a data reference request or update request on the request signal, the server 1 in the cluster can accept the data reference request or update request from the outside. In the present embodiment, as will be described later, an example in which an HTTP GET method or a PUT method is used as a request will be described.

  Regarding the point where the data storage destination moves within the cluster of problem (2), it is necessary to follow the configuration change of the data distribution adopted in the cluster. Therefore, the cluster data storage / management method is, for example, the KVS (Key Value Store) mechanism, and the distribution algorithm (for example, the consistent hash method) is used. Note that KVS is a data storage / management technique in which a corresponding unique indicator (key) is set for arbitrarily stored data (value: value), and the key and value are stored in pairs. In this way, even if the data storage destination is changed within the cluster due to redundancy, etc., it is possible to follow the server 1 that is the data storage destination by specifying the key. It becomes.

  Next, as a premise for explaining the details of this embodiment, main examples of cache data reference processing and update processing by the server 1 (1x) in the client cluster 20 will be described with reference to FIGS. explain. The detailed processing flow of the reference processing and update processing in the data cache system 100 will be described later with reference to FIGS. 11 to 15 including other examples.

  FIG. 3 shows an example in which one server 1 (1x) in the client cluster 20 processes a cache data reference request. Specifically, in the server 1 (1x), the cache manager 113 (FIG. 5) that manages the cache data from the application (application processing unit 112 shown in FIG. 5: expressed as “APL” in FIG. 3) issues a reference request. The case where it received is shown. Here, the application processing unit 112 is set to execute processing via the cache manager 113 when referring to or updating data.

  First, the cache manager 113 of the server 1 (1x) receives a data reference request from the application processing unit 112 (step S10). Note that a key (data key) for identifying the data is attached to the reference request.

  When the cache manager 113 receives the reference request, the cache manager 113 searches its own storage unit 13 (FIG. 5: local cache) based on the data key, and returns the data to the application processing unit 112 if the data is stored. (Step S11).

  If the cache manager 113 does not store the data in its own storage unit 13 (local cache), the cache manager 113 acquires the data with read authority from another cache manager 113 having write authority (step S12). Then, the cache manager 113 returns the acquired data to the application processing unit 112 (step S13).

  Here, the Read authority is an authority to read (refer) data, and refers to an authority to refer to the data, and data cannot be written. The write authority is an authority to write (update) data and to update the data. When the cache manager 113 has the Write authority, it has the Read authority at the same time, and can grant the Read authority to other cache managers 113 or transfer the Write authority.

  The cache data stores data (data value) with the data key and metadata of the data attached (for example, see reference numeral 1000 in FIG. 11). Including the identifier (eg, “p1”) of the cache manager 113 including the identifier (eg, “p1 @ XXX”) of the cache manager 113 including the identifier of the cache manager 113 including the identifier (eg, “q”). The key of the cache manager (for example, “q @ YYY”) is described. Note that “XXX” and “YYY” are addresses (for example, IP addresses) of the load balancer B serving as the entrance of the cluster, and indicate the cluster to which the cache manager 113 belongs. The cache manager 113 having the write authority updates this metadata when granting the read authority to another cache manager or transferring the write authority.

  FIG. 4 shows an example in which one server 1 (1x) in the client cluster 20 processes a cache data update request. Specifically, in the server 1 (1x), the cache manager 113 receives an update request from the application processing unit 112.

  First, the cache manager 113 of the server 1 (1x) receives a data update request from the application processing unit 112 (step S20). This update request is accompanied by a key of the data (data key).

  Upon receiving the update request, the cache manager 113 searches its own storage unit 13 (local cache) based on the data key, stores the data, and refers to the metadata attached to the data. If the user has Write permission, the data is updated. Then, the cache manager 113 transmits an update notification of the data to the other application processing unit 112 that caches the data (step S21).

On the other hand, if the cache manager 113 does not store the data in its own storage unit 13 (local cache), or if it does not have the write authority, the other cache manager 113 having the write authority for the data does not know. Then, by transferring the data update request, the other cache manager 113 having the write authority is made to update the data.
In addition, the cache manager 113 does not store the data in its own storage unit 13 (local cache) or does not have the write authority even if it is stored. Is updated with the write authority from another cache manager 113 having the write authority for the data (step S22).
Then, the cache manager 113 updates the data and transmits an update notification of the data to the other application processing unit 112 that caches the data (step S23).

  As described above, when the cache manager 113 having the write authority is updated with the data or the metadata attached to the data, the cache manager 113 transmits an update notification to the other application processing unit 112 that caches the data. By doing so, the coherence (consistency) of the data in the cluster is maintained.

  In the data cache system 100 according to the present embodiment, as described above, the data key is attached to the data to be cached, and together with the data key, the cache manager 113 having the read authority and the write authority for the data. Metadata indicating a key (cache manager key) is attached to the data and stored as cache data. When a reference request or an update request is transmitted from outside to the cluster, by setting a request option in an existing request signal (for example, a request by an HTTP GET method or a PUT method), a data key, a transmission source, Send the destination cache manager key and the processing details. The node in the cluster that has received the request signal can identify the node in charge of the data based on the data key attached to the request signal or the cache manager key and receive the request signal.

<Server configuration>
Hereinafter, the server 1 (1x, 1y) constituting the cluster (client cluster 20 and original data cluster 10) according to the present embodiment will be specifically described. As shown in FIG. 2, the server 1 will be described as having both the function of the dispatcher D and the function of the processor P. In addition, here, the server 1 (1y) constituting the original data cluster 10 and the server 1 (1x) constituting the client cluster 20 have the same functions, and therefore will be described as the server 1.

FIG. 5 is a functional block diagram illustrating a configuration example of the server 1 according to the present embodiment.
As shown in FIG. 2, the server 1 is communicably connected to the load balancer B, and is also communicably connected to other servers 1 other than itself constituting the cluster. Then, a request such as a reference request or an update request is received from the outside via the load balancer B.
As shown in FIG. 5, the server 1 includes a control unit 11, an input / output unit 12, and a storage unit 13.

  The input / output unit 12 inputs / outputs information to / from the load balancer B and other servers 1 other than itself. The input / output unit 12 includes a communication interface that transmits and receives information via a communication line, and an input / output interface that performs input / output between an input unit such as a keyboard (not shown) and an output unit such as a monitor. Consists of

  The control unit 11 controls the entire server 1 and includes an information receiving unit 111, an application processing unit 112, a cache manager 113, a distribution processing unit 114, and an information transmission unit 115. The control unit 11 is realized, for example, by a CPU (Central Processing Unit) developing and executing a program stored in the storage unit 13 of the server 1 on a RAM (Random Access Memory) or the like.

  The information receiving unit 111 receives information from the load balancer B and other servers 1 in the cluster via the input / output unit 12.

  The application processing unit 112 executes processing of an application for providing a service to a client. When using the cache data stored in the storage unit 13, the application processing unit 112 outputs a request such as a data reference request or an update request to the cache manager 113. This request includes a key (data key) for uniquely identifying the data.

The cache manager 113 performs overall control of processing related to cache data.
Specifically, when the cache manager 113 receives a data reference request from the application processing unit 112 and stores the data in its own storage unit 13 (cache data storage unit 131), the data Is returned to the application processing unit 112.
On the other hand, when the cache manager 113 does not store the data in its own storage unit 13 (cache data storage unit 131), the cache manager 113 receives data with read authority from another cache manager 113 having write authority for the data. The data is acquired and stored in the storage unit 13 (cache data storage unit 131), and then the data is returned to the application processing unit 112. Details of this processing will be described later with reference to FIG.

In addition, when the cache manager 113 receives a data update request from the application processing unit 112, the cache manager 113 stores the data in its own storage unit 13 (cache data storage unit 131), and also has write authority for the data. If it has, the data is updated to new data based on the update request.
On the other hand, the cache manager 113 does not store the data in its own storage unit 13 (cache data storage unit 131) or stores the data (that is, has the Read right). When the user does not have the write authority, the data is updated by receiving the write authority from the application processing unit 112 of the other server 1 having the write authority. At this time, the cache manager 113 can update the data by requesting the other server having the write authority to update the data. Details of this processing will be described later with reference to FIGS.

  The cache manager 113 searches whether or not the data is stored in the storage unit 13 (cache data storage unit 131) using the data key attached to the received request. Further, it is determined whether or not the user has write permission for the data with reference to the metadata attached to the data. Details of the processing of the cache manager 113 will be described later with reference to FIGS.

  Here, the data key in this embodiment is described as “A @ YYY”, for example. Here, “A” is an identifier (data identifier) unique to the data. “YYY” is the address of the load balancer B that is the entrance (external transmission destination) of the cluster where the original data of the data is stored. As described above, the data key is assumed to be a combination of the data identifier and the address of the load balancer B unique to the original data cluster 10. Therefore, as the information of the data key, the original of the data “A” is sent to any one of the plurality of servers 1 (1y) constituting the original data cluster 10 including the load balancer B (address: “YYY”). It means that data is always stored.

  When the cache manager 113 transmits a request to a cluster other than the cluster to which the cache manager 113 belongs, the cache manager 113 transmits the request as a request including at least the cache manager key (cache manager key) and the data key. In the present embodiment, an example in which request options are set in the HTTP GET method and the PUT method will be described later. However, the request options are not limited to HTTP as long as similar request options can be set.

  When the information receiving unit 111 receives a request from the outside, the distribution processing unit 114 performs a predetermined distribution algorithm using a KVS mechanism (here, a consistent hash method and a data key attached to the request). The server 1 in charge of storing and managing the data (specifically, the cache manager 113 of the server 1) is determined.

  Here, as described above, the cache manager 113 of each server 1 is given a unique identifier in the cluster. In the consistent hash method, both the data key and the identifier of the cache manager 113 are Find the value (hash value) multiplied by the hash function. Then, a value (hash value) obtained by applying each identifier of the cache manager 113 to the hash function is placed on the closed hash space, and a value (hash value) obtained by applying the data key attached to the request to the hash function is, for example, a hash space. It is assumed that the cache manager 113 that meets first in the clockwise direction is in charge of the data.

  Specifically, each server 1 includes cache data management information 132 in the storage unit 13 as common information in the cluster, and the distribution processing unit 114 manages the data based on the cache data management information 132. The cache manager 113 to be determined can be determined, and the address of the server including the cache manager 113 can be known.

  FIG. 6 is a diagram showing a data configuration of the cache data management information 132 according to the present embodiment. As shown in FIG. 6, the cache data management information 132 stores an identifier of the cache manager 113, a hash value that is a value obtained by multiplying the identifier by a hash function, and an address of a server provided with the cache manager 113.

For example, in FIG. 6, for the data whose value (hash value) obtained by applying the data key to the hash function is “0” to “56”, the cache of the server whose address is “xx0” shown in the first line The manager 113 (cache manager identifier “p0”) is in charge. Similarly, for the data in which the value obtained by applying the data key to the hash function (hash value) is “57” to “172” obtained by adding 1 to “56”, the address indicated by the second line is “xx1”. The server's cache manager 113 (cache manager identifier “p1”) is in charge.
In this way, based on the cache data management information 132, the data and the cache manager 113 in charge of the data are associated with each other.
In the present embodiment, it is assumed that one server 1 has one cache manager 113. However, it is possible to set so that a plurality of cache managers 113 exist in one server 1. Further, the function of the distribution processing unit 114 corresponds to the function of the dispatcher D shown in FIG.

  Returning to FIG. 5, the information transmission unit 115 inputs information generated by the cache manager 113 or the like (requests such as reference requests and update requests and response information thereof) to other servers 1 and load balancer B other than itself. Transmit via the output unit 12.

  The storage unit 13 includes a storage device such as a RAM, a hard disk, and a flash memory, and stores the cache data management information 132 (FIG. 6) and the like. Further, the storage unit 13 includes a cache data storage unit 131 (also referred to as a local cache) as an area for storing data as cache data in a RAM or the like.

<Processing flow of data cache system>
Next, a processing flow of the data cache system 100 according to the present embodiment will be described.
First, the processing of the cache manager 113 of the server 1 will be described in detail with reference to FIGS. 7 to 10, and then the overall processing flow of the data cache system 100 will be described with reference to FIGS. 11 to 15. To do.

≪Process of cache manager≫
[Cache data reference processing]
FIG. 7 is a flowchart showing processing when the cache manager 113 of the server 1 (1x) of the client cluster 20 according to the present embodiment receives a request for a data reference request from the application processing unit 112.

  First, the cache manager 113 (identifier “p”) receives a data reference request from the application processing unit 112 (step S100). As shown in FIG. 7, “p.get (key, cachetype)” indicates that the cache manager 113 (cache manager identifier “p”) uses data “key” (data key) and “cachetype” as arguments. This means that the get method is executed as a data reference request. “Cachetype” is attached with “Read cache” requesting Read authority (reference authority) or “Write cache” requesting update authority.

  Next, the cache manager 113 executes data acquisition processing from its own cache data storage unit 131 (local cache) using a key (data key), here, for example, “A @ YYY” (step S101). ).

Here, the cache manager 113 determines whether there is no corresponding data in the cache data storage unit 131 (local cache) or whether the corresponding data is invalidated (step S102). Then, if there is no corresponding data or does not correspond to invalidation (step S102 → false), that is, if the corresponding data is found, the cache manager 113 uses the data (here, “v”) as application processing. It returns to the part 112 (step S103), and a process is complete | finished.
On the other hand, if there is no corresponding data or invalidation (step S102 → true), the cache manager 113 proceeds to the next step S104.

In step S104, the cache manager 113 determines whether or not the own server is the original server of the key (data key). Here, the original server constitutes the original data cluster 10 including the address of the load balancer B, which is indicated by “YYY” in the data key (A @ YYY) and becomes the entrance of the cluster (external transmission destination). Server 1 (1y). When the cache manager 113 is the original server of the key (step S104 → true), the cache manager 113 returns “null” indicating that the data cannot be acquired to the application processing unit 112 (step S105). Finish the process.
On the other hand, if the cache manager 113 is not the original server of the key (step S104 → false), the process proceeds to the next step S106.

  In step S106, the cache manager 113 determines the cache type of the reference request. Then, the cache manager 113 proceeds to step S107 if the cachetype is Read cache, and proceeds to step S108 if the cachetype is Write cache.

In step S107, the cache manager 113 notifies the original server of the key that the server 1 (self server) additionally has the Read authority (“ADD_R” method). In other words, the cache manager 113 requests the original server of the key to send the data after adding the cache manager key to the read authority of the metadata of the data (giving Read authority). To do.
The cache manager 113 notifies this to the server 1 (1y) constituting the original data cluster 10, but the destination is the address (YYY) of the load balancer B of the original data cluster 10. Specifically, the cache manager 113 executes “v = p.send (ADD_R, key, key.location (), null)” of the send method, for example, as a request by the HTTP GET method,
http: //YYY/A.xml? sourcecachemanager = p @ XXX & cachetype = read
Execute.
In this request, “YYY” is the address (YYY) of the load balancer B of the original data cluster 10, “A” is the data identifier, and the part after the “?” Mark is the request option. Become. In this request option part, the cache manager key (“sourcecachemanager = p @ XXX”) of the cache manager 113 that is the transmission source, the request for read authority as the processing content (“cachetype = read”), etc. are described. .

On the other hand, in step S108, the cache manager 113 notifies the original server of key that the server 1 (self server) has the write authority (“ADD_W” method). That is, the cache manager 113 requests the original server of key to transmit the data after rewriting the write authority of the metadata of the data to the own server (with the write authority).
Specifically, the cache manager 113 executes the send method “v = p.send (ADD_W, key, key.location (), null)”, for example, as a request by an HTTP GET method,
http: //YYY/A.xml? sourcecachemanager = p @ XXX & cachetype = write
Execute.

  When the processing of step S107 and step S108 is completed, in step S109, the cache manager 113 receives the corresponding data from the server 1 having the write authority. Then, the cache manager 113 stores the data in the cache data storage unit 131 (local cache) (“p.put_local (key, value = v)”).

  Subsequently, assuming that the corresponding data is found, the cache manager 113 returns the data (here, “v”) to the application processing unit 112 (step S110), and ends the processing.

[Cache data update processing]
8 and 9 show the case where the cache manager 113 of the server 1 (1x) of the client cluster 20 according to the present embodiment receives a request for an update request for data (here, data “w”) from the application processing unit 112. It is a flowchart which shows the process of.

  First, the cache manager 113 (identifier “p”) receives a data update request from the application processing unit 112 (step S200). In response to this update request, the cache manager 113 executes the put method with the data “key” (data key) and “value” (updated data) as arguments (p.put (key, value)).

  In the next step S201, the cache manager 113 executes data acquisition processing using its key (data key) from its own cache data storage unit 131 (local cache) (w = p.get_local (key)). .

  Here, the cache manager 113 determines whether there is no corresponding data in the cache data storage unit 131 (local cache) or whether the corresponding data is invalidated (step S202). Then, if there is no corresponding data or does not correspond to invalidation (step S202 → false), that is, if the corresponding data is found, the cache manager 113 proceeds to step S203.

  In step S203, the cache manager 113 acquires a storage location having a write authority from the metadata attached to the corresponding data. Here, the storage location having the write authority means, for example, the data “q @ YYY” of the <Write cache> tag in the metadata (see reference numeral 1000) shown in FIG. That is, among the servers 1 (1y) of the cluster (original data cluster 10) indicated by the address “YYY” of the load balancer B, the cache manager 113 whose cache manager identifier is “q” has the write authority. Yes. As described above, “q @ YYY” is set as the key of the cache manager 113.

  Next, it is determined whether or not the cache manager 113 has write authority (step S204). If the cache manager 113 itself has the Write authority (step S204 → true), the process proceeds to the next step S208. On the other hand, if it is not the cache manager 113 with write authority (step S204 → false), the process proceeds to step S205, and the cache manager 113 with write authority is requested to update (p.send (PUT, key, q, value )).

On the other hand, if the corresponding data is not found in the cache data storage unit 131 (local cache) in step S202 (step S202 → true), it is determined whether or not the own server is the original server of the key (data key) ( Step S206). If the cache manager 113 is not the original server of the key (step S206 → false), in step S207, the cache manager 113 requests the original server of the key to update (p.send (PUT, key, key.location ( ), value)). The cache manager 113 makes this request to the server 1 (1y) constituting the original data cluster 10, but the destination is the address (YYY) of the load balancer B of the original data cluster 10.
On the other hand, if the cache manager 113 is the original server of the key (step S206 → true), the process proceeds to step S208.

  In step S208, the cache manager 113 updates the data (cache data) stored in its own cache data storage unit 131 (local cache) (p.put_local (key, value)). At this time, the cache manager 113 also updates the metadata of the data if necessary.

  Next, the process proceeds to FIG. In FIG. 9, since the data has been updated in step S <b> 208 of FIG. 8, the cache manager 113 invalidates the data before update stored in the other cache manager 113, and Read authority and Write for the updated data. A process of transmitting update data to another authorized cache manager 113 is executed.

  First, in step S209, the cache manager 113 acquires all the cache managers 113 having the Read authority from the metadata before updating the data, and generates a list.

  Subsequently, the cache manager 113 performs settings for invalidating the data before update (step S210), that is, generates an invalid notice.

  Then, the cache manager 113 determines whether or not all the cache managers 113 in the list generated in step S209 have been processed (step S211). If all the cache managers 113 having the Read authority have not been processed (step S211 → false), the process proceeds to the next step S212.

  The cache manager 113 sets the storage location r (data of <Read cache> tag of metadata) of the i-th (initial value i = 1) cache manager 113 in the list (step S212), and the cache manager 113 sets the cache location in step S213. An invalid notification is transmitted to the manager 113 (p.send (PUT_L, key, r, w)). In step S214, the cache manager 113 adds 1 to [i] and returns to step S211.

  When the cache manager 113 completes the processing of all the cache managers 113 having the Read authority (Step S211 → true), the process proceeds to Step S215, and the cache manager 113 having the Read authority and the Write authority is updated from the updated metadata. Get all and generate a list.

  Next, the cache manager 113 determines whether or not all the cache managers 113 in the list generated in step S215 have been processed (step S216), and if all the cache managers 113 have not been processed (step S216). S216 → false), the process proceeds to the next step S217.

  The cache manager 113 sets a storage location r (data of <Read cache> tag and <Write cache> tag of metadata) of the i-th (initial value i = 1) cache manager 113 in the list (step S217). In step S218, data (update data) as update contents is transmitted to the cache manager 113 (p.send (PUT_L, key, r, value)). In step S219, the cache manager 113 adds 1 to [i] and returns to step S216.

  In step S216, if all the cache managers 113 in the list generated in step S215 have been processed (step S216 → true), the process ends.

[Cache Manager receive method processing]
Next, a process in which a cache manager 113 of a server 1 receives information by a send method executed by another cache manager 113 and executes a receive method will be described.
FIG. 10 is a flowchart showing a flow of processing in which the cache manager 113 according to the present embodiment executes the reception method.

  First, in step S300, the cache manager 113 of the server 1 receives the information transmitted when the other cache manager 113 executes the send method via the information receiving unit 111, and receives the receive method. Is executed (p.receive (method, key, from, value)).

  Next, the cache manager 113 determines a method of the received information (step S301). If the method of the received information is “PUT”, the cache manager 113 proceeds to step S302, and if it is “PUT_L”, the process proceeds to step S303, where “ADD_R” (adding Read authority) or “ADD_W” ( If the write authority is transferred), the process proceeds to step S304.

  In step S302, the cache manager 113 determines that the data update request has been received, assuming that the method is “PUT”, and executes the cache data update process shown in FIGS. p.put (key, value)).

  In step S303, the cache manager 113 assumes that the method is “PUT_L” and updates the data (cache data) stored in its own cache data storage unit 131 (local cache) (p. 30). put_local (key, value)).

  In step S304, when the method is “ADD_R” or “ADD_W”, first, the cache manager 113 executes acquisition processing of the corresponding data from its own cache data storage unit 131 (local cache) ( w = p.get_local (key)).

  Next, in step S305, the cache manager 113 acquires a storage location (data of <Write cache> tag of metadata) having the write authority from the metadata attached to the corresponding data (q = w.get_metadata (write_cache) )).

  Subsequently, the cache manager 113 determines again the method of the received information (step S306). The cache manager 113 proceeds to step S307 if the method is “ADD_R”, and proceeds to step S308 if the method is “ADD_W”.

  In step S307, the cache manager 113 generates information in which Read authority is set in the metadata of the data (w.set_metadata (read_cache, from)).

  In step S308, the cache manager 113 generates information in which write authority is set in the metadata of the data (w.set_matadata (write_cache, from)).

Subsequent to step S307 and step S308, the cache manager 113 determines whether or not the cache manager 113 itself has the write authority (step S309).
If the cache manager 113 itself is not the write authority (step S309 → false), in step S310, the cache manager 113 requests the cache manager 113 having the write authority to update (p.send (PUT, key , q, w)).
On the other hand, if the cache manager 113 has write authority (step S309 → true), in step S311, the cache manager 113 stores data (cache data) stored in its own cache data storage unit 131 (local cache). ) Is updated (p.put_local (key, value)).

  When the processes of steps S310 and S311 are completed, the cache manager 113 transmits data “w” to the cache manager 113 that has transmitted the send method (step S312).

<< Overall Processing Flow of Data Cache System 100 >>
Next, the flow of the cache data reference processing and update processing in the entire data cache system 100 (between the client cluster 20 and the original data cluster 10) according to the present embodiment will be described with reference to FIGS. I will explain.

[Cache data reference processing]
FIG. 11 shows an example in which the cache manager 113 of the server 1 (1x) constituting the client cluster 20 acquires data with a Read right when receiving a reference request without storing the corresponding data of the request. Show.

  First, the cache manager 113 (identifier “p1”) of the client cluster 20 receives a reference request from the application processing unit 112 (step S30). It is assumed that “A @ YYY” is attached to the reference request as a key (data key) of data requesting reference.

In step S31, the cache manager 113 that has received the reference request searches its own cache data storage unit 131 (local cache), and does not store the corresponding data. Therefore, the cache manager 113 executes the send method (p1). .send (ADD_R, A @ YYY, YYY, null)), a data acquisition request is transmitted to the original data cluster 10. For example, the cache manager 113 executes an HTTP GET method, sets a request option, and transmits a data acquisition request (http: //YYY/A.xml? Sourcecachemanager = p1 @ XXX & cachetype = read).
As a result, the cache manager 113 transmits a request to the load balancer B (IP: YYY) serving as the entrance of the original data cluster 10, and the load balancer B distributes the request to an arbitrary server 1 (1y). Then, based on the data key (A @ YYY), the distribution processing unit 114 of the distributed server 1 (1y) detects the responsibility for storing the data as original data with reference to the cache data management information 132. The request is transmitted to the server 1 (1y) in charge of the data (here, the cache manager 113 with the identifier “q”).

Through the above processing, the cache manager 113 (identifier “q”) in charge of the original data receives the request (step S32), and executes the receive method (q.receive (ADD_R, A @ YYY, p1 @ XXX, null)).
Then, if the cache manager 113 (identifier “q”) in charge of the original data has the write authority, in step S33, the read authority (<read cache> tag) of the metadata of the data is set to “p1 @ “XXX” is added (see reference numeral 1000 in FIG. 11), and the file α including the data is transmitted to the client cluster 20. Here, for example, the cache manager 113 executes the HTTP PUT method and transmits the data (http: //XXX/A.xml? Destinationcachemanager = p1 @ XXX & sourcecachemanager = q @ YYY).
Note that in the description of the <read cache> tag denoted by reference numeral 1000 in FIG. 11, the description of the end tag that should be <read cache> p1 @ XXX </ read cache> is omitted as </>. Similarly, the end tag of <write cache> in FIG. 11, and also in FIGS. 12 to 15, the end tag is similarly omitted as </>.
By the above processing, the cache manager 113 transmits a request to the load balancer B (IP: XXX) serving as the entrance of the client cluster 20, and the load balancer B distributes the request to an arbitrary server 1 (1x). Then, the distribution processing unit 114 of the distributed server 1 (1x) extracts the cache manager 113 (cache manager identifier “p1”) as the destination, refers to the cache data management information 132, and sends the request to the data The data is transmitted to the server 1 (1x) in charge (the cache manager 113 with the identifier “p1”).

  If the cache manager 113 (identifier “q”) that is in charge of the original data does not have the write authority, the cache manager 113 refers to the metadata of the data and writes the write authority (<write cache> tag). The request for the acquisition request is further transmitted to the cache manager 113 described in FIG.

  As a result of the above processing, the cache manager 113 (identifier “p1”) of the client cluster 20 that has received the data stores the data in its own cache data storage unit 131 (local cache) (step S34). Return to the application processing unit 112.

  In this way, when the cache manager 113 of the server 1 (1x) constituting the client cluster 20 receives the reference request without storing the corresponding data of the request, the data is acquired with the Read authority. can do.

[Cache data reference processing (with Write permission)]
FIG. 12 shows an example in which the cache manager 113 of the server 1 (1x) constituting the client cluster 20 acquires data with a write authority when receiving a reference request in a state where the corresponding data of the request is not stored. Show.

  First, the cache manager 113 (identifier “p1”) of the client cluster 20 receives a reference request including the write authority from the application processing unit 112 (step S40). It is assumed that “A @ YYY” is attached to the reference request as a key (data key) of data requesting reference.

  In step S41, the cache manager 113 that has received the reference request including the write authority searches its own cache data storage unit 131 (local cache), and executes the send method because the corresponding data is not stored. (P1.send (ADD_W, A @ YYY, YYY, null)), and sends a data acquisition request to the original data cluster 10 (http: //YYY/A.xml? Sourcecachemanager = p1 @ XXX & cachetype = write).

  The cache manager 113 (identifier “q”) in charge of the original data identified by the processing of the distribution processing unit 114 of the arbitrary server 1 (1y) via the load balancer B (IP: YYY) of the original data cluster 10 ) Receives a request (step S42), executes a receive method (q.receive (ADD_W, A @ YYY, p1 @ XXX, null)).

  Next, this cache manager 113 (identifier “q”) having the write authority writes “p1 @ XXX” in the write authority (<write cache> tag) of the metadata of the data in step S43, and reads “Q @ YYY”, which is its own cache manager key, is added to the authority (<read cache> tag) as the charge of the original data (see reference numeral 1001 in FIG. 12). This is because the cache manager 113 that manages the original data retains the Read authority even when the Write authority is transferred. Then, the cache manager 113 transmits the file α including the data to the client cluster 20 (http: //XXX/A.xml? Destinationcachemanager = p1 @ XXX & sourcecachemanager = q @ YYY).

  The cache manager 113 (identifier “p1”) that has received data with write authority from any server 1 (1x) via the load balancer B (IP: XXX) of the client cluster 20 sends the data to its own The data is stored in the cache data storage unit 131 (local cache) (step S44), and the data is returned to the application processing unit 112.

In this way, when the cache manager 113 of the server 1 (1x) constituting the client cluster 20 receives the reference request without storing the corresponding data of the request, the data is acquired with the Write authority. can do.
In addition, since the cache manager 113 on the client cluster 20 side has the write authority and can update the cache data by itself, the cache hit effect on the frequently rewritten data can be enhanced.

[Cache data update processing]
FIG. 13 shows the cache manager 113 (identifier “p2”) when the cache manager 113 (identifier “p2”) has a write authority for data (data key “A @ YYY”) and receives an update request. p2 ") itself shows an example of updating this data.

  First, the cache manager 113 (identifier “p2”) of the client cluster 20 receives a data update request from the application processing unit 112 (step S50). It is assumed that “A @ YYY” is attached to the update request as a key (data key) of data requesting update.

  The cache manager 113 (identifier “p2”) searches its own cache data storage unit 131 (local cache), the corresponding data exists, and by referring to the metadata of the data, the cache manager 113 (identifier “p2”) The data is confirmed and the data is updated (step S51).

  Here, the cache manager 113 (identifier “p2”) refers to the read authority of the metadata before update, and transmits an invalid notice to the cache manager 113 having the read authority before update (step S52). Here, as indicated by reference numeral 1002, since the cache manager 113 of the client cluster 20 (identifier “p1” and the cache manager 113 of the original data cluster 10 (identifier “q”) has the read authority before the update, An invalid notification of data (“α”) is transmitted to the two cache managers 113. For example, for the cache manager 113 (identifier “q”) of the original data cluster 10, “http: // YYY / A.xml? Destinationcachemanager = q @ YYY & sourcecachemanager = p2 @ XXX & invalidate = true ”is sent.

  Subsequently, the cache manager 113 (identifier “p2”) refers to the read authority of the updated metadata, and transmits the update data to the cache manager 113 having the updated read authority (step S53). Here, as indicated by reference numeral 1003, since the cache manager 113 (identifier “q”) of the original data cluster 10 has the updated Read authority, the update data is transmitted to the cache manager 113. Note that the cache manager 113 (identifier “p1”) of the client cluster 20 does not have the Read authority in the updated metadata, and therefore the update data is not transmitted.

  The cache manager (identifier “q”) that has received the update data from any server 1 (1y) via the load balancer B (IP: YYY) of the original data cluster 10 has its own cache data storage unit 131 (local The update data (“α ′”) is stored in the cache, thereby updating the data (step S54).

  In this way, when the cache manager 113 of the client cluster 20 has an authority to write data and receives an update request, the cache manager 113 itself can update the data. In addition, an invalidity notification is transmitted to the other cache manager 113 that stores the data before update, and the updated data is transmitted to the other cache manager 113 based on the updated metadata. Cache data coherence can be maintained.

[Cache data update request]
FIG. 14 shows an example in which when the cache manager 113 having no write authority wants to update data, the cache manager 113 having the write authority is requested to update.

  First, the cache manager 113 (identifier “p1”) of the client cluster 20 receives a data update request from the application processing unit 112 (step S60). It is assumed that “A @ YYY” is attached to the update request as a key (data key) of data requesting update.

  Here, in step S61, the cache manager 113 (identifier “p1”) searches its own cache data storage unit 131 (local cache) and does not store the corresponding data, and executes the send method. (P1.send (PUT, A @ YYY, YYY, α ')) and a data update request is transmitted to the load balancer B (IP: YYY) of the original data cluster 10 (http: //YYY/A.xml ? sourcecachemanager = p1 @ XXX).

The cache manager 113 (identifier “q”) in charge of the original data identified by the processing of the distribution processing unit 114 of the arbitrary server 1 (1y) via the load balancer B (IP: YYY) of the original data cluster 10 ) Receives a data update request in step S62 and executes a receive method (q.receive (PUT, A @ YYY, p1 @ XXX, α ')), Update the data.
Here, if the cache manager 113 (identifier “q”) that is responsible for the original data does not have the write authority, the cache manager 113 refers to the metadata of the data and writes the write authority (<write cache Further, an update request is transmitted to the cache manager 113 described in <Tag>.

  The cache manager 113 (identifier “q”) having the write authority adds the cache manager (identifier “p1”) to the read authority (<read cache> tag) of the metadata of the data (see reference numeral 1004), and reads the authority. Then, the update data is transmitted to the cache manager 113 (identifier “p1”) of the client cluster 20 (step S63). At this time, if another cache manager 113 is registered in the read authority (<read cache> tag) of the metadata of the data, the update data is also transmitted to the cache manager at the same time.

  The cache manager 113 (cache manager identifier “p1”) that has received the update data from any server 1 (1x) via the load balancer B (IP: XXX) of the client cluster 20 sends the data to its own The data is stored in the cache data storage unit 131 (local cache) (step S64).

  By doing so, when the cache manager 113 without the write authority wants to update the data, the cache manager 113 with the write authority is requested to update the data, and after the data is updated, the cache manager 113 itself reads. Can have authority.

[Update notification only with the cluster IP address]
FIG. 15 omits the identifier of the cache manager 113 and reads only the cluster address (actually, the address of the load balancer B that is the entrance of the cluster) in the metadata read authority (<read cache> tag). Shows the flow of update notification.

  Here, the data is updated by the cache manager 113 (identifier “q”) having the write authority of the original data cluster 10 (step S70), and the load balancer of the client cluster 20 is added to the metadata read authority (<readcache> tag). Only the IP address (XXX) of B is described (see reference numeral 1005). In this case, the cache manager 113 (identifier “q”) having the write authority transmits an update notification to all the cache managers 113 of the client cluster 20 in step S71 (http: //XXX/A.xml). ? destinationcachemanager = p * @ XXX & sourcecachemanager = q @ YYY).

  The load balancer B of the client cluster 20 that has received the update notification randomly distributes the update notification to some server 1 (1x) (step S72). Then, the server 1 (1x) to which the update notification is distributed transmits an update notification to each server 1 (1x) using a mechanism that broadcasts to the entire cluster (step S73).

  In this way, each server 1 in the cluster can have read authority and store data.

  As described above, according to the data cache system 100 according to the present embodiment, the cache data information is updated while the cluster configurations of the caller (client) and the callee (original data storage server) are hidden from each other. It is possible to notify each other. Even if the configuration of the server 1 in the cluster changes due to failure or expansion, the request is sent to the correct storage destination of the reference target or update target data, and the reference or update processing is performed. It can be executed and maintain cluster coherence.

1, 1x, 1y server 10 original data cluster 11 control unit 12 input / output unit 13 storage unit 20 client cluster 30 client terminal 40 external terminal 100 data cache system 111 information receiving unit 112 application processing unit 113 cache manager 114 distribution processing unit 115 information Transmission unit 131 Cache data storage unit (local cache)
132 Cache data management information B Load balancer D (D ₁ , D ₂ , D ₃ ) Dispatcher P (P ₁ , P ₂ , P ₃ ) Processor

Claims (4)

A client cluster in which a plurality of servers that store data as cache data constitutes a cluster and an original data cluster in which a plurality of servers that store the original data of the cache data constitute a cluster are connected via a network A data cache system,
A first server that is a server included in the client cluster, and a second server that is a server included in the original data cluster,
(1) A data key composed of an identifier unique to the data and an address of the original data cluster for storing the original data of the data, an identifier unique to a cache manager that executes read processing and write processing of the data, and Using a cache manager key composed of an address of the cluster indicating the client cluster or the original data cluster to which the cache manager belongs, metadata storing a cache manager having the right to read and write the data, (2) a storage unit that stores the data key, and cache data management information in which the data key is associated with each unique identifier of each of the plurality of cache managers set in the servers in each cluster;
A data processing unit that extracts the data key attached to the data, refers to the cache data management information, and determines the cache manager corresponding to the data;
The cache manager,
The cache manager of the first server is
When the data reference request is received, the data key attached to the reference request is searched for whether the data is stored in the storage unit of the data, and the data is stored. If not, the address of the original data cluster is extracted from the data key, and an acquisition request to which the data key of the data and the cache manager key of the cache manager are attached is a first request message to the original data cluster. And send it in
The distribution processing unit of the second server extracts the data key from the received first request message and refers to the cache data management information for the cache manager that stores the original data corresponding to the data. Sending the first request message to the determined cache manager;
The cache manager of the second server that has received the first request message refers to the metadata of the data, and when the cache manager is authorized to write, the cache manager key is obtained from the first request message. Extracting and attaching the extracted cache manager key to the metadata read authority of the data, extracting the address of the client cluster from the cache manager key, and transferring the data to the cache manager of the first server Send
The cache manager of the first server receives the data, stores it in its storage unit, and outputs the data. The cache manager of the second server that has received the first request message refers to the metadata of the data and extracts the cache manager having the write authority when the cache manager does not have the write authority. The data cache system according to claim 1, wherein the first request message is transmitted to the cache manager having the extracted write authority. The cache manager of the first server is
When the data update request is received, the data key attached to the update request is searched for whether or not the data is stored in the storage unit of the data, and the data is stored. If the original data cluster address is extracted from the data key when the metadata does not have write authority with reference to the metadata even if stored, the data key of the data and the cache manager An update request with a cache manager key is sent in the second request message to the original data cluster,
The distribution processing unit of the second server extracts the data key from the received second request message and refers to the cache data management information for the cache manager that stores the original data corresponding to the data. Sending the second request message to the determined cache manager;
The cache manager of the second server that has received the second request message refers to the metadata of the data, and updates the data when the user has write permission, and the second request message The cache manager key is extracted from the metadata, and the extracted cache manager key is added to the read permission of the metadata. Then, the address of the client cluster is extracted from the cache manager key, and the updated data is stored in the first data. To the server's cache manager,
The data cache system according to claim 1 or 2, wherein the cache manager of the first server receives the updated data, stores the updated data in its own storage unit, and outputs the data. The cache manager of the second server that has received the second request message refers to the metadata of the data and extracts the cache manager having the write authority when the cache manager does not have the write authority. The data cache system according to claim 3, wherein the second request message is transmitted to a cache manager having the extracted write authority.

Images