JP2013205962A - Subsystem control node, control method for subsystem, program, and service provision system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a subsystem control node which effectively utilizes resources,and enhances capability of an entire system.SOLUTION: A subsystem control node includes: an access distribution estimation part for estimating a distribution of accesses to cache data from cache use state data including information on access to cache data stored by a distribution cache cluster; a resource increase/decrease influence determination part for calculating from the estimated access distribution, a resource increase/decrease influence degree indicative of an influence on a service provision system including the distribution cache cluster caused by the resource increase/decrease in the distribution cache cluster; a resource increase/decrease object calculation part for determining on the basis of the resource increase/decrease degree whether the resource included in the distribution cache cluster should be increased or decreased, or whether a resource included in another subsystem should be increased or decreased; and a control instruction part for instructing a subsystem determined by the resource increase/decrease object calculation part to increase or decrease of a resource.

Description

  The present invention relates to a subsystem control node, a subsystem control method, a program, and a service providing system. In particular, the present invention relates to a subsystem control node including a distributed cache cluster, a subsystem control method, a program, and a service providing system.

  There is a system that includes a plurality of subsystems and provides services to clients. For example, a shopping system configured on the WEB (World Wide Web) is applicable. Such a system includes an application server cluster that includes a plurality of application servers that receive requests from clients and perform processing based on the received requests.

  Also, the above-mentioned system often includes a distributed cache cluster as a subsystem that caches part of the processing contents performed on a specific application server and reuses it on another application server. In such a distributed cache cluster, the read cache of the data stored in the database server and the cache of the calculation result of the application server are held in the distributed cache server which is the configuration node, and can be used by other application servers. .

  Here, Patent Document 1 discloses a technique for identifying an area where access is concentrated by monitoring access to a disk storage and caching the area.

  Also, in Patent Document 2, is the I / O performance from the access source reduced based on I / O (Input / Output) status information that is information indicating the input / output status accompanying the input / output command from the access source? A technique for determining whether or not to use the cache and changing the usage state of the cache so as to improve the I / O performance is disclosed.

  Furthermore, Patent Document 3 discloses a technique for adjusting the number of servers in a cluster system according to an increase or decrease in service demand. The technique disclosed in Patent Document 3 uses a cluster reconfiguration technique to increase or decrease the number of computers constituting the system.

JP 2010-002942 A JP 2010-286923 A JP 2005-092662 A

  Each disclosure of the above prior art document is incorporated herein by reference. The following analysis was made by the present inventors.

  As described above, a system that provides a specific service is configured by a plurality of subsystems, and a distributed cache cluster may be included as a subsystem. A distributed cache cluster can increase the capacity of the distributed cache cluster by increasing the resources contained therein.

  However, even if the capacity of the distributed cache cluster is increased, it may not contribute to improving the performance of the entire service providing system. In such a case, it is not meaningful to allocate resources (computer nodes) to the distributed cache cluster, and it is desirable to take measures such as not allocating resources to the distributed cache cluster or allocating them to other subsystems.

  However, in a system composed of a plurality of subsystems, indices such as which of the configuration nodes included in the subsystem should be increased or decreased with respect to load fluctuations with the passage of time include Patent Documents 1 to 3. There is no disclosure in the prior art. Therefore, even if resources are allocated to the distributed cache cluster, it is assumed that the allocated resources are not effectively used and do not contribute to the performance improvement of the entire system.

  From the above, a subsystem control node, a subsystem control method, a program, and a service providing system that effectively utilize resources and improve the capability of the entire service providing system are desired. Details of the capacity enhancement of the distributed cache cluster and the influence on the performance of the entire service providing system will be described later.

  According to the first aspect of the present invention, an access distribution estimation unit that estimates an access distribution for the cache data from cache usage data including information related to access to the cache data stored in the distributed cache cluster, and the estimated access Based on the resource increase / decrease influence degree, a resource increase / decrease influence determination unit that calculates a resource increase / decrease influence degree indicating the influence of the increase / decrease of the resource included in the distributed cache cluster on the service providing system including the distributed cache cluster from the distribution; A resource increase / decrease target calculation unit that determines whether to increase / decrease resources included in the distributed cache cluster or increase / decrease resources included in another subsystem, and a subsystem determined by the resource increase / decrease target calculation unit Increase of resources Subsystem control node and a control instruction unit for instructing a is provided.

  According to the second aspect of the present invention, an access distribution estimation step of estimating an access distribution for the cache data from cache usage status data including information related to access to the cache data stored in the distributed cache cluster, and the estimated access From the distribution, a resource increase / decrease influence determination step for calculating a resource increase / decrease influence degree indicating an influence of increase / decrease of resources included in the distributed cache cluster on a service providing system including the distributed cache cluster, and based on the resource increase / decrease influence degree A resource increase / decrease target calculation step for determining whether to increase / decrease resources included in the distributed cache cluster or increase / decrease resources included in another subsystem, and the subsystem determined in the resource increase / decrease target calculation step Against The method of subsystems including a step of instructing the resource of increase or decrease, is provided.

According to the third aspect of the present invention, an access distribution estimation process for estimating an access distribution for the cache data from cache usage status data including information on access to the cache data stored in the distributed cache cluster, and the estimated access Based on the resource increase / decrease influence determination processing for calculating the resource increase / decrease influence degree indicating the influence of the increase / decrease of the resource included in the distributed cache cluster on the service providing system including the distributed cache cluster from the distribution, , A resource increase / decrease target calculation process for determining whether to increase / decrease resources included in the distributed cache cluster or increase / decrease resources included in other subsystems, and the subsystem determined in the resource increase / decrease target calculation process Against Program for executing a process of instructing the resource of increase or decrease, to a computer to control the sub-system control node is provided.
This program can be recorded on a computer-readable storage medium. The storage medium can be non-transient such as a semiconductor memory, a hard disk, a magnetic recording medium, an optical recording medium, or the like. The present invention can also be embodied as a computer program product.

  According to a fourth aspect of the present invention, a distributed cache comprising: a distributed cache node capable of storing cache data; and a cache usage status monitoring unit that generates cache usage status data including information related to access to the cache data. A cluster, an access distribution estimation unit that estimates an access distribution for the cache data from the cache usage status data, and the distributed cache cluster when the resources included in the distributed cache cluster are increased or decreased from the estimated access distribution A resource increase / decrease influence determining unit that calculates an influence degree of resource increase / decrease indicating an influence on a system including the resource increase / decrease effect, or increase / decrease resources included in the distributed cache cluster based on the resource increase / decrease influence degree Included resource A subsystem control node comprising: a resource increase / decrease target calculation unit that determines whether to increase / decrease, and a control instruction unit that instructs the subsystem determined by the resource increase / decrease target calculation unit to increase / decrease resources A service providing system is provided.

  According to each aspect of the present invention, there are provided a subsystem control node, a subsystem control method, a program, and a service providing system that effectively use resources and improve the capability of the entire service providing system.

It is a figure for demonstrating the outline | summary of one Embodiment. It is a figure which shows an example of the whole structure of the service provision system. It is a figure which shows an example of the whole structure of the service provision system 8 which concerns on 1st Embodiment. It is a flowchart which shows an example of operation | movement of the control instruction node 6a. It is a figure which shows an example of the numerical data of an access condition. It is a figure which shows an example of the access distribution graph calculated using the parameter of the estimated model function. It is a figure which shows an example of the access distribution graph calculated using the parameter of the estimated model function. It is a figure which shows an example of the access distribution graph calculated using the parameter of the estimated model function.

  First, an outline of an embodiment will be described with reference to FIG. Note that the reference numerals of the drawings attached to the outline are attached to the respective elements for convenience as an example for facilitating understanding, and the description of the outline is not intended to be any limitation.

  As described above, there is a clear indicator regarding which resources included in a subsystem should be increased or decreased in response to load fluctuations over time in a system composed of multiple subsystems. do not do. For this reason, even if resources are allocated to the distributed cache cluster, the allocated resources may not be used effectively, and may not contribute to improving the performance of the entire system. In view of the above, a subsystem control node that effectively utilizes resources and improves the capabilities of the entire service providing system is desired.

  Therefore, the subsystem control node 100 shown in FIG. 1 is provided as an example. The subsystem control node 100 includes an access distribution estimation unit 101 that estimates an access distribution for cache data from cache usage state data including information related to access to cache data stored in the distributed cache cluster, and a distributed distribution from the estimated access distribution. Included in the distributed cache cluster based on the resource increase / decrease influence determining unit 102 that calculates the resource increase / decrease influence degree indicating the influence of the increase / decrease of the resource included in the cache cluster on the service providing system including the distributed cache cluster The resource increase / decrease target calculation unit 103 that determines whether to increase or decrease the resources to be increased or decreased and the resources included in other subsystems, and the subsystem determined by the resource increase / decrease target calculation unit 103 to the resource And a control instructing unit 104 for instructing the increase or decrease, the.

  The subsystem control node 100 estimates the access distribution from the access information (cache usage status data) for the cache data, and when the capacity of the distributed cache cluster is increased or decreased (the resources are increased or decreased) from the estimated access distribution, the performance of the entire system is improved. The resource increase / decrease influence level is calculated by quantifying how much it contributes to. This resource increase / decrease influence level is compared with an arbitrarily specified threshold, and only when the improvement in the performance of the entire system can be expected due to the increase in the resources included in the distributed cache cluster, the resources included in the distributed cache cluster are increased. As a result, resources can be used effectively and the capability of the entire service providing system can be improved.

  Furthermore, the following forms are possible.

  [Mode 1] As in the subsystem control node according to the first aspect.

  [Mode 2] The resource increase / decrease influence determining unit is configured to model access to the cache data when it is assumed that the access to the cache data follows a zip distribution, and the number of caches that can be stored in the distributed cache cluster. , It is preferable to calculate the resource increase / decrease influence degree.

  [Mode 3] When the resource increase / decrease target calculation unit receives an instruction to increase the resource of the subsystem constituting the service providing system, and the resource increase / decrease influence degree is equal to or greater than a predetermined threshold, When the instruction to increase the resource included in the distributed cache cluster and decrease the resource of the subsystem constituting the service providing system is received, and when the influence of the resource increase / decrease is smaller than a predetermined threshold, It is preferable to reduce the resources included in the cache cluster.

  [Mode 4] Preferably, the resource increase / decrease target calculation unit determines increase / decrease in resources included in a subsystem different from the distributed cache cluster based on resource load information of a subsystem different from the distributed cache cluster.

  [Mode 5] The subsystem control method according to the second aspect.

  [Mode 6] The resource increase / decrease influence determining step includes a model function that models access to the cache data when it is assumed that the access to the cache data follows a zip distribution, and the number of caches that can be stored in the distributed cache cluster. , It is preferable to calculate the resource increase / decrease influence degree.

  [Mode 7] When the resource increase / decrease target calculation step receives an instruction to increase the resource of the subsystem constituting the service providing system, and the resource increase / decrease influence degree is equal to or greater than a predetermined threshold, When the instruction to increase the resource included in the distributed cache cluster and decrease the resource of the subsystem constituting the service providing system is received, and when the influence of the resource increase / decrease is smaller than a predetermined threshold, It is preferable to reduce the resources included in the cache cluster.

  [Mode 8] Preferably, the resource increase / decrease target calculation step determines increase / decrease in resources included in a subsystem different from the distributed cache cluster based on resource load information of a subsystem different from the distributed cache cluster.

  [Mode 9] The program according to the third aspect.

  [Mode 10] The resource increase / decrease influence determination process includes a model function that models access to the cache data when it is assumed that access to the cache data follows a zip distribution, and the number of caches that can be stored in the distributed cache cluster. , It is preferable to calculate the resource increase / decrease influence degree.

  [Mode 11] When the resource increase / decrease target calculation process receives an instruction to increase the resource of a subsystem constituting the service providing system, and the resource increase / decrease influence degree is equal to or greater than a predetermined threshold, When the instruction to increase the resource included in the distributed cache cluster and decrease the resource of the subsystem constituting the service providing system is received, and when the influence of the resource increase / decrease is smaller than a predetermined threshold, It is preferable to reduce the resources included in the cache cluster.

  [Mode 12] Preferably, the resource increase / decrease target calculation processing determines increase / decrease in resources included in a subsystem different from the distributed cache cluster based on resource load information of a subsystem different from the distributed cache cluster.

  [Mode 13] The service providing system according to the fourth aspect.

  Hereinafter, specific embodiments will be described in more detail with reference to the drawings.

  First, the service providing system 1 including a distributed cache cluster will be described.

  FIG. 2 is a diagram illustrating an example of the overall configuration of the service providing system 1.

  The service providing system 1 includes an application server cluster 2, a distributed cache cluster 3, a distributed storage system 4, a resource pool 5, a control instruction node 6, and a control event notification unit 7. Each subsystem of the application server cluster 2, the distributed cache cluster 3, and the distributed storage system 4 includes a configuration node (not shown).

  The service providing system 1 provides an arbitrary service by connecting the application server cluster 2, the distributed cache cluster 3, and the distributed storage system 4 via a network. For example, in a system realized by a three-layer model such as a WEB shopping system, a UI (presentation layer) that provides a WEB screen exists in addition to the subsystem shown in FIG. 2 and has a function of generating dynamic content. The application server cluster 2 is in charge.

  In addition, the distributed cache cluster 3 is responsible for caching and speeding up the contents of the generated dynamic content and the static content of the presentation layer.

  Furthermore, the distributed storage system 4 is in charge of the function of a database (data layer) that stores information for generating static contents and information for generating dynamic contents.

  The application server cluster 2 accepts a processing request from an upper layer that provides a UI (User Interface) function of a service to be provided, accesses data stored in the distributed storage system 4 as necessary, and uses the data Execute the calculation process. At that time, the application server cluster 2 responds to the same request, data that needs to be temporarily stored (for example, data that is not placed in a shopping cart of the WEB shopping system), distributed storage system 4 Are stored in the distributed cache cluster 3.

  Further, the application server cluster 2 returns a calculation result to the issuer of the request. In addition, the application server cluster 2 uses the data held in the distributed cache cluster 3 for a request that only has to return an equivalent result, without performing recalculation processing or accessing the distributed storage system 4. And responds to access.

  The distributed cache cluster 3 temporarily holds data of the distributed storage system 4 accessed by the application server cluster 2 and data generated by calculation by the application server cluster 2. The distributed cache cluster 3 responds to access requests from other systems (subsystems) including the application server cluster 2 using temporarily held data. Note that cache data is often stored in the memory of a constituent node from the viewpoint of performing a response to a request at high speed.

  Since the distributed cache cluster 3 exists, the application server cluster 2 and the like do not need to repeatedly access the distributed storage system 4 or execute the calculation with the same calculation result. As a result, the load on the application server cluster 2 and the distributed storage system 4 can be reduced, and improvement in the performance of the service providing system 1 as a whole and improvement in resource utilization can be expected.

  The distributed storage system 4 holds data necessary for the service to be provided and responds to a request from the application server cluster 2 or the like. The distributed storage system 4 is often configured to include a database server, a file system server, and the like.

  Here, the application server cluster 2, the distributed cache cluster 3, and the distributed storage system 4 are each configured to include a single or a plurality of computer nodes. This computer node corresponds to a resource necessary for realizing the function of each subsystem. As the computer node, a server including a CPU (Central Processing Unit), a memory, an HDD, and the like can be assumed. Alternatively, it may be a virtual machine that emulates a plurality of computers on one computer.

  The resource pool 5 is a control node that holds and manages increase / decrease of resources allocated to the application server cluster 2 and the like. More specifically, the resource pool 5 is a control node that executes a program that realizes a function of allocating or detaching (removing) resources from or to a subsystem such as the application server cluster 2. As shown in FIG. 2, the resource pool 5 may be managed centrally or may be managed for each subsystem (a resource pool may exist for each subsystem).

  The control instruction node 6 instructs increase / decrease of resources included in each subsystem. The control event notification unit 7 notifies the control instruction node 6 of a trigger for executing the resource increase / decrease process. For example, an instruction from a system administrator or the like is applicable.

  Next, the performance of the service providing system 1 will be described.

  By using the distributed cache cluster 3, it is possible to improve the performance of the reference processing request in the service to be provided.

  On the other hand, access to the distributed storage system 4 is required for update processing requests, and even if the capacity of the distributed cache cluster 3 is increased (although many resources are allocated), the entire service providing system 1 The performance is not always improved. For example, when an update process request is received, if the request is a process in which read access to cache data held by the distributed cache cluster 3 is dominant, an update request is received. However, the performance improvement of the entire system can be expected. Alternatively, when an update processing request writes data to the distributed cache cluster 3 once and writes data to the distributed storage system 4 asynchronously when data is written to the distributed storage system 4, The contribution of the distributed cache cluster 3 to the request can be expected.

  However, except for these cases, it cannot be expected to contribute much to the performance improvement of the entire service providing system 1 by increasing the capacity of the distributed cache cluster 3 with respect to update processing requests.

  Resources managed by the resource pool 5 can be assigned to any of the application server cluster 2, the distributed cache cluster 3, and the distributed storage system 4. In the case of such a service providing system 1, when load requests for services increase, it becomes a problem to which subsystem (application server cluster 2 or the like) a resource should be allocated.

  At that time, it can be considered that a large number of resources are simply allocated to a subsystem that lacks resources. However, in such a correspondence, in a system such as the service providing system 1, the operation of each subsystem affects each other, so the performance of the entire system does not necessarily improve. More specifically, even if many resources are allocated to the distributed cache cluster 3, if the load requests that the distributed cache cluster 3 can respond to have not increased, it does not contribute to the performance improvement of the service providing system 1 as a whole. In other words, there is no disclosure in the prior art including Patent Documents 1 to 3 in what ratio the finite resources should be allocated to each subsystem.

  Furthermore, when the load of the service providing system 1 decreases, it is required to stop surplus resources. For example, in order to reduce the power cost, power supply to some resources is stopped. Alternatively, in the case where the cost is determined in proportion to the usage amount of resources represented by cloud computing, the service provider (system operator) needs to operate with the resources reduced as much as possible.

  Next, when the resources of the distributed cache cluster 3 are increased or decreased in the service providing system 1, the influence on the entire system is analyzed.

  First, consider a case where the resources included in the distributed cache cluster 3 are reduced. In this case, this means that the cache capacity that can be provided by the distributed cache cluster 3 is reduced, and therefore the types of data that are not cached increase. Accordingly, the cache data corresponding to the reduced capacity is responded not by cache access but by recalculation by the application server cluster 2 itself. Therefore, the load on the application server cluster 2 increases. Furthermore, the access load to the distributed storage system 4 increases as the recalculation process increases.

  Next, consider a case where the resources included in the distributed cache cluster 3 are increased. In this case, since the capacity for caching the calculation results of the application server cluster 2 increases, the calculation processing by the application server cluster 2 decreases. Therefore, the access load on the distributed storage system 4 is also reduced.

  At this time, if the access distribution to the distributed storage system 4 is a distribution such as a zip distribution, it is unlikely that data with a low access frequency order will be accessed again. For example, the access frequency to the WEB page often follows a zip distribution.

  Since data with low access frequency rank is unlikely to be accessed again, even if the calculation result is cached in the distributed cache cluster 3, such data is rarely accessed again. Therefore, the load on the application server cluster 2 and the distributed storage system 4 is not reduced. That is, in such a case, even if the capacity of the distributed cache cluster 3 is increased, the performance improvement of the entire service providing system 1 cannot be expected.

  From the above, in order to effectively use limited resources, it is required to appropriately determine whether to increase the resources included in the distributed cache cluster 3 or to increase the resources included in other subsystems. .

[First Embodiment]
The first embodiment will be described in more detail with reference to the drawings.

  FIG. 3 is a diagram illustrating an example of the overall configuration of the service providing system 8 according to the present embodiment. 3, the same components as those in FIG. 2 are denoted by the same reference numerals, and the description thereof is omitted.

  First, details of each component included in the service providing system 8 will be described.

  The distributed cache cluster 3a includes a cache usage status monitoring unit 10, a node increase / decrease unit 11, and distributed cache nodes 12 and 13.

  The distributed cache nodes 12 and 13 are resources for realizing the distributed cache cluster 3a. In FIG. 3, the distributed cache cluster 3a is realized by two nodes. Of course, one or more arbitrary nodes can be used. The distributed cache nodes 12 and 13 include a means for realizing a distributed cache and a cache data storage means (not shown).

  The cache usage status monitoring unit 10 acquires the access count of data registered as a cache in the distributed cache cluster 3a. Further, the cache usage status monitoring unit 10 responds with the access count of the registered data in response to a request from the access distribution estimation unit 40 included in the control instruction node 6a.

  Various forms of implementation of the cache usage status monitoring unit 10 are possible. For example, each time the distributed cache nodes 12 and 13 are used, a usage history is counted, and such usage history is collected by a program operating on an arbitrary computer node, and is stored by storage management means such as a database program. Can be considered.

  The node increasing / decreasing unit 11 has a function of increasing or decreasing the number of distributed cache nodes constituting the distributed cache cluster 3a. The node increase / decrease unit 11 leaves the resource allocated to the distributed cache cluster 3a from the distributed cache cluster 3a and notifies the resource pool 5 to that effect. Alternatively, when increasing the number of allocated nodes, resources are secured from the resource pool 5 and participate in the distributed cache cluster 3a.

  The node increasing / decreasing unit 11 can be realized by executing a program in any one of the distributed cache nodes constituting the distributed cache cluster 3a, or can be realized by operating the programs of a plurality of nodes in cooperation.

  When the node increase / decrease unit 11 decreases the number of distributed cache nodes in the distributed cache cluster 3a, the cache data is transferred to another distributed cache node according to the number of accesses acquired by the cache usage monitoring unit 10 (Migration) can also be performed. For example, when the number of distributed cache nodes is decreased, the cache data may be rearranged so that cache data having a large number of accesses remains in the cache, and then the process of leaving the distributed cache node may be performed.

  Further, such data relocation is not performed, the cache data held by the detached distributed cache node is deleted from the distributed cache cluster 3a, and the deleted cache data is recalculated by the application server cluster 2. Alternatively, it may be acquired by re-accessing the distributed storage system 4.

  The application server cluster 2 includes configuration nodes 21 and 22 and a node increase / decrease unit 20.

  The configuration nodes 21 and 22 are resources for realizing functions required for the application server cluster 2. In FIG. 3, the application server cluster 2 is realized by two nodes, but it is needless to say that one or more arbitrary nodes can be used.

  The node increasing / decreasing unit 20 has a function of increasing / decreasing the number of constituent nodes 21 and 22 constituting the application server cluster 2. Similar to the node increase / decrease unit 11, the node increase / decrease unit 20 secures resources from the resource pool 5 and assigns them to the application server cluster 2, or leaves the resources from the application server cluster 2 and returns them to the resource pool 5.

  The distributed storage system 4 holds data necessary for the service to be provided and responds to a request from the application server cluster 2 or the like. The distributed storage system 4 is often configured to include a database server, a file system server, and the like.

  There is also a distributed storage technology that realizes a system that stores or uses data by using a computer node HDD (Hard Disk Drive) or memory by connecting a large number of computer nodes via a network. In such a distributed storage technology, functions such as which computer node data is allocated to and which computer node performs processing are realized by a program or hardware. In the distributed storage system 4, by dynamically changing the operation according to the system state, the resource usage in the system is adjusted, and the performance improvement for the system user (client computer) is realized. The service providing system 8 shown in FIG. 3 uses a distributed storage system 4 that employs such a distributed storage technology. However, if the resources of the distributed storage system 4 are not controlled (the number of computer nodes is increased or decreased), storage systems and database systems that do not employ the distributed storage technology can be included in the distributed storage system 4.

  Furthermore, the same applicant has proposed a storage system that realizes suppression of power consumption by shifting some of the constituent nodes included in the distributed storage system to the low power consumption mode (Japanese Patent Application: Japanese Patent Application: Patent Application). International application PCT / JP2011 / 071857 based on priority claim based on 2010-224697). In the service providing system 8 according to the present embodiment, such a system can be used as the distributed storage system 4.

  The distributed storage system 4 includes configuration nodes 31 and 32 and a node increase / decrease unit 30. The node increase / decrease unit 30 may execute control for transitioning to a low power consumption mode while causing some nodes of the distributed storage system 4 to participate in the storage system.

  The control instruction node 6a includes an access distribution estimation unit 40, a resource increase / decrease influence determination unit 41, a resource increase / decrease target calculation unit 42, and a control instruction unit 43. Each of these units is realized by a computer program or hardware. The control instruction node 6a corresponds to the subsystem control node 100 described above. The operation of each unit included in the control instruction node 6a will be described later.

  The control event notification unit 7 notifies the control instruction node 6a of a trigger for executing the resource increase / decrease determination process. There are various possible implementations, and they are listed below.

  First, it is conceivable to use a timer operated by the same computer as the control instruction node 6a or by another computer. That is, the control event notification unit 7 is realized as a module that periodically starts the determination process.

  Secondly, it is conceivable to use a user interface such as a button assigned to a screen or the like monitored by the system administrator or a command executed from the command line.

  Third, an instruction from an external cluster management system or the like via an API (Application Program Interface) can be considered.

  Fourth, when an indication of resource increase / decrease is detected by an external load monitoring system or load prediction system, an instruction via an API or the like can be considered.

  Fifth, when an external power monitoring system or the like detects the necessity for a resource increase / decrease request, an instruction via an API or the like can be considered.

  Sixth, an instruction via an API or the like is conceivable when any resource (configuration node) of a subsystem that realizes a service is monitored and a bottleneck or surplus is detected. More specifically, the bottleneck or surplus at that time is detected by a method in which the CPU usage rate of the computer is higher or lower than a threshold value for a certain period of time.

  Seventh, it is conceivable that a system or an administrator detects and notifies a change in a load pattern for a service.

  The trigger for executing the above resource increase / decrease determination processing can be regarded as a resource adjustment request for the system that implements the service. Hereinafter, it is assumed that the resource adjustment request information includes designation as to whether to increase or decrease the resource. However, it goes without saying that the increase or decrease in resources may be determined inside the control instruction node 6a.

  Next, operations of the service providing system 8 and the control instruction node 6a will be described.

  FIG. 4 is a flowchart showing an example of the operation of the control instruction node 6a.

  In step S01, the control instruction node 6a accepts a control event notification issued by the control event notification unit 7. The mode of control event notification is as described above.

  In step S02, the access distribution estimation unit 40 acquires the cache usage status data from the cache usage status monitoring unit 10, and outputs the cache usage status data to the resource increase / decrease influence determination unit 41.

  In step S03, the resource increase / decrease influence determination unit 41 estimates the access distribution to the cache from the received cache usage status data.

  Here, the access distribution estimation method in this step will be described in detail. There are various models of access distribution probability models (for example, uniform distribution, zip distribution, etc.). The resource increase / decrease influence determination unit 41 estimates the access distribution using these distribution models. This access distribution often has a different distribution model depending on the service (application) to be provided. Even in the same service (application), the load pattern may differ depending on the time and time.

  FIG. 5 is a diagram illustrating an example of numerical data of the access status. The cache key in FIG. 5 is a unique key for the application server cluster 2 (or another subsystem) to access the cache data held by the distributed cache cluster 3a. The cache data associated with this cache key is stored and held in the distributed cache cluster 3a and becomes accessible. For example, in the case of a service that responds by calculating some data for each user, a user ID or the like is used as the cache key. In the example shown in FIG. 5, the usage count of the distributed cache cluster 3a is stored and managed for each cache key, and used as cache usage status data.

  The cache usage status monitoring unit 10 responds with data as shown in FIG. 5 as cache usage status data. The number of accesses is desirably the number of accesses within a predetermined period. This is because, except for special cases, extremely old cache data is considered invalid data.

なお、Ｎは、全要素の総数、ｋはアクセス順位とする。また、モデル関数を指数分布（ｙ＝ａ×ｅ^ｂｘ）と仮定すれば、ａ及びｂの値を推定する。 The resource increase / decrease influence determination unit 41 uses the cache usage status data (access distribution data) and the model function, and estimates the parameter of the model function by approximation using a mathematical method or nonlinear regression analysis. For example, assuming a zip distribution as the distribution model, the value of the parameter s can be estimated by the following formula (1).

N is the total number of all elements, and k is the access order. If the model function is assumed to be an exponential distribution (y = a × e ^bx ), the values of a and b are estimated.

  In step S04, the resource increase / decrease influence determination unit 41 uses the estimated model function (access distribution) to calculate a score that quantifies the effect when the resource is increased or decreased, and uses the calculated score as the resource increase / decrease target calculation unit 42. Output to. This score corresponds to the resource increase / decrease influence degree described above.

  At this time, if the control event notification (resource adjustment request) is an instruction to increase the resource, the score when the resource included in the distributed cache cluster 3a is increased is calculated. On the other hand, if the control event notification is an instruction to decrease the resource, a score when the resource included in the distributed cache cluster 3a is decreased is calculated. In the following description, the control event notification is an instruction to increase resources.

  Note that if the control event notification does not clarify the instruction regarding the increase or decrease of the resource, the increase or decrease of the resource is determined in this step. Various methods can be considered as the determination method at that time. For example, the monitoring information of the resource usage status of each subsystem is acquired, or the information monitoring the power usage status is acquired, and the SLA (Service Level Agreement) such as the response time of the realized service is acquired, The determination is based on these.

  Next, score calculation in the resource increase / decrease influence determination unit 41 will be described in detail.

  6 to 8 are diagrams showing examples of access distribution graphs calculated using the parameters of the estimated model function. 6 to 8, the horizontal axis indicates the access amount ranking of requests, and the vertical axis indicates the number of accesses.

  In this case, the capacity of the cache data that can be held by the distributed cache cluster 3a can be handled in the same manner as the width on the horizontal axis of each figure. For example, assuming that the cache providing capacity per distributed cache node (resource) included in the distributed cache cluster 3a is 100 MB and the cache data per cache key is 1 MB on average, 100 caches can be stored in one distributed cache node. It is a piece. In other words, the 100 access ranks on the horizontal axis in each figure can be held as a cache.

  Accordingly, if the distributed cache cluster 3a operates with an algorithm that holds cache data in the order of the most accesses, the distributed cache cluster 3a operates with three distributed cache nodes and can hold 100 caches per unit. Then, the cache data held by the distributed cache cluster 3a is cache data having an access order of 1 to 300. The calculation of the cache capacity is only an outline and is strictly different. However, it is not necessary to calculate the cache capacity strictly in the description of the present embodiment.

  Further, since the first access is an access for registering a cache, it does not contribute to reducing the load on the application server cluster 2 and the distributed storage system 4. That is, the access in which the distributed cache cluster 3a contributes to the performance improvement of the entire service providing system 8 is an area 200 in FIG. Therefore, the area of this region can be regarded as a contribution to improving the performance of the distributed cache cluster 3a with respect to the entire system.

  Further, when one distributed cache node is added, the contribution to the performance improvement of the entire system is an area 201 in FIG. That is, the area of the region 201 is equivalent to the degree of influence on the performance improvement of the entire system due to the increase of the distributed cache nodes. The area of this region 201 is treated as a score in this embodiment.

なお、ｆ（ｘ）は分布モデル関数、ｘはアクセス順位、αは構成ノード１台あたりに保持可能なキャッシュキーの数、ｎは現在の分散キャッシュクラスタ３ａに含まれる構成ノードの台数である。同様に、構成ノードを１台離脱させた場合についても同様に算出することができる。 If it is assumed that the cache capacities held by the respective cache keys are substantially equal, a score corresponding to this area can be calculated from the following equation (2).

Note that f (x) is a distribution model function, x is an access order, α is the number of cache keys that can be held per configuration node, and n is the number of configuration nodes included in the current distributed cache cluster 3a. Similarly, the same calculation can be performed when one component node is detached.

  In step S05, the resource increase / decrease target calculation unit 42 compares the calculated score with a predetermined threshold value.

  When the calculated score is equal to or greater than the threshold (step S05, Yes branch), the target for increasing the resource is the distributed cache cluster 3a (step S06). On the other hand, if the threshold value is not exceeded (step S05, No branch), the target whose resource is to be increased is set as another subsystem (application server cluster 2 or distributed storage system 4) (step S07).

  Here, when the process proceeds to step S07, the resources included in the other subsystem (application server cluster 2 or distributed storage system 4) are increased instead of the distributed cache cluster 3a. Any method can be adopted as a method for selecting which subsystem resource to increase. As an example, the following method can be considered.

  The resource load information (for example, CPU usage rate or memory usage) of the application server cluster 2 or the distributed storage system 4 is compared with a threshold, and resources are allocated to subsystems exceeding the threshold. Furthermore, when both resource load information exceeds a threshold value, a resource is allocated to a more tight subsystem.

  In step S01, when an instruction to reduce resources is accepted as a control event notification, the branch destination differs depending on the determination result in step S05. More specifically, when the calculated score is equal to or greater than the threshold, the resources of other subsystems are reduced, and when the calculated score is not exceeded, the resources of the distributed cache cluster 3a are reduced.

  In addition, the system administrator may determine a threshold value to be compared with the score by an operation IF (not shown) or other program in the control instruction node 6a. Furthermore, the threshold value may be adjusted based on the past control history.

  In step S08, the resource increase / decrease target calculation unit 42 notifies the control instruction unit 43 of the subsystem to be increased / decreased in the resource determined in step S06 or step S07. The control instruction unit 43 issues a resource increase / decrease request to the node increase / decrease unit of the target subsystem (distributed cache cluster 3a, application server cluster 2 or distributed storage system 4).

  The control instruction node 6a according to the present embodiment is a distributed control in which a plurality of systems in an infrastructure for realizing cloud computing are mixed, resource control of each layer of the system architecture based on the three-layer model, increase / decrease of resources included in the distributed cache cluster, and cloud computing It can be applied to control each resource of the system.

  As described above, in the service providing system 8 according to the present embodiment, the cache usage status monitoring unit 10 monitors the cache usage status of the distributed cache cluster 3a, and the access distribution estimation unit 40 accesses the cache data from the cache usage status data. Estimate the distribution. Further, the resource increase / decrease influence determination unit 41 calculates a score based on the estimated access distribution, and selects a subsystem that is a target for increasing / decreasing the resource based on the calculated score. In particular, the influence on the entire system due to the increase / decrease of the distributed cache nodes included in the distributed cache cluster 3a is estimated, and the subsystem to be the resource allocation destination is determined using the estimation result.

  More specifically, using the estimated access distribution, the effect on the entire system when the capacity of the distributed cache cluster 3a is enhanced is quantified (score is calculated), and it is worth increasing resources (distributed cache nodes). Only if you increase resources. On the other hand, even if the capacity of the distributed cache cluster 3a is increased, if the contribution to the performance improvement of the entire system cannot be expected, the resources of other subsystems are increased. As a result, resources can be added to a subsystem that has a high effect of increasing resources, so that system operation with high resource utilization efficiency can be realized.

  Furthermore, the power consumption of the entire service providing system 8 can be suppressed. The load distribution for the subsystem is calculated from the request distribution, the resources necessary for each subsystem are calculated, and the resources are reduced according to the calculation result. By operating such a system, the power consumption of the entire service providing system 8 can be suppressed. That is, the power consumption of the service providing system 8 can be suppressed by effectively using the resources.

  Each disclosure of the cited patent documents and the like cited above is incorporated herein by reference. Within the scope of the entire disclosure (including claims) of the present invention, the embodiments and examples can be changed and adjusted based on the basic technical concept. Various disclosed elements (including each element of each claim, each element of each embodiment or example, each element of each drawing, etc.) within the scope of the claims of the present invention, Selection is possible. That is, the present invention of course includes various variations and modifications that could be made by those skilled in the art according to the entire disclosure including the claims and the technical idea.

DESCRIPTION OF SYMBOLS 1, 8 Service provision system 2 Application server cluster 3, 3a Distributed cache cluster 4 Distributed storage system 5 Resource pool 6, 6a Control instruction node 7 Control event notification part 10 Cache utilization condition monitoring part 11, 20, 30 Node increase / decrease part 12, 13 Distributed cache nodes 21, 22, 31, 32 Configuration nodes 40, 101 Access distribution estimation unit 41, 102 Resource increase / decrease influence determination unit 42, 103 Resource increase / decrease target calculation unit 43, 104 Control instruction unit 100 Subsystem control nodes 200, 201 region

Claims (10)

An access distribution estimation unit that estimates an access distribution to the cache data from cache usage data including information about access to the cache data stored in the distributed cache cluster;
From the estimated access distribution, a resource increase / decrease influence determination unit that calculates a resource increase / decrease influence degree indicating an influence of increase / decrease of resources included in the distributed cache cluster on a service providing system including the distributed cache cluster;
A resource increase / decrease target calculation unit that determines whether to increase or decrease the resources included in the distributed cache cluster or increase or decrease the resources included in other subsystems based on the resource increase / decrease influence degree;
A control instruction unit for instructing increase / decrease of resources to the subsystem determined by the resource increase / decrease target calculation unit;
A subsystem control node comprising:   The resource increase / decrease influence determination unit is based on a model function that models access to the cache data when it is assumed that access to the cache data follows a zip distribution, and the number of caches that can be stored in the distributed cache cluster. The subsystem control node according to claim 1, which calculates a resource increase / decrease influence degree. When the resource increase / decrease target calculation unit receives an instruction to increase the resource of a subsystem constituting the service providing system and the resource increase / decrease influence degree is equal to or greater than a predetermined threshold, the resource increase / decrease target calculation unit Increase the resources included,
The resource included in the distributed cache cluster is reduced when an instruction to reduce the resource of a subsystem constituting the service providing system is received, and when the resource increase / decrease influence degree is smaller than a predetermined threshold. 1 or 2 subsystem control nodes.   4. The resource increase / decrease target calculation unit determines increase / decrease in resources included in a subsystem different from the distributed cache cluster based on resource load information of a subsystem different from the distributed cache cluster. The subsystem control node according to 1. An access distribution estimating step for estimating an access distribution for the cache data from cache usage data including information on access to the cache data stored in the distributed cache cluster;
From the estimated access distribution, a resource increase / decrease influence determination step for calculating a resource increase / decrease influence degree indicating an influence of increase / decrease of resources included in the distributed cache cluster on a service providing system including the distributed cache cluster;
A resource increase / decrease target calculation step for determining whether to increase / decrease the resources included in the distributed cache cluster or increase / decrease the resources included in other subsystems based on the resource increase / decrease influence degree;
Instructing the increase / decrease of resources to the subsystem determined in the resource increase / decrease target calculation step;
A method for controlling a subsystem, comprising:   The resource increase / decrease influence determining step is based on a model function that models access to the cache data when access to the cache data follows a zip distribution, and the number of caches that can be stored in the distributed cache cluster. 6. The subsystem control method according to claim 5, wherein a resource increase / decrease influence degree is calculated. In the resource increase / decrease target calculation step, when an instruction to increase the resource of the subsystem constituting the service providing system is received and the influence of the resource increase / decrease is equal to or greater than a predetermined threshold, Increase the resources included,
The resource included in the distributed cache cluster is reduced when an instruction to reduce the resource of a subsystem constituting the service providing system is received, and when the resource increase / decrease influence degree is smaller than a predetermined threshold. 5 or 6 subsystem control method.   The resource increase / decrease target calculation step determines increase / decrease in resources included in a subsystem different from the distributed cache cluster based on resource load information of a subsystem different from the distributed cache cluster. A subsystem control method according to claim 1. An access distribution estimation process for estimating an access distribution for the cache data from cache usage data including information on access to the cache data stored in the distributed cache cluster;
From the estimated access distribution, a resource increase / decrease influence determination process for calculating a resource increase / decrease influence degree indicating an influence of increase / decrease of resources included in the distributed cache cluster on a service providing system including the distributed cache cluster;
Resource increase / decrease target calculation processing for determining whether to increase / decrease the resources included in the distributed cache cluster or increase / decrease the resources included in other subsystems based on the resource increase / decrease influence degree;
A process for instructing the subsystem determined in the resource increase / decrease target calculation process to increase / decrease resources;
A program that causes a computer that controls the subsystem control node to execute. A distributed cache node capable of storing cache data; and
A cache usage status monitoring unit that generates cache usage status data including information relating to access to the cache data;
A distributed cache cluster comprising:
An access distribution estimation unit that estimates an access distribution for the cache data from the cache usage data;
A resource increase / decrease influence determining unit that calculates a resource increase / decrease influence degree indicating an influence on a system including the distributed cache cluster when the resources included in the distributed cache cluster are increased / decreased from the estimated access distribution;
A resource increase / decrease target calculation unit that determines whether to increase or decrease the resources included in the distributed cache cluster or increase or decrease the resources included in other subsystems based on the resource increase / decrease influence degree;
A control instruction unit for instructing increase / decrease of resources to the subsystem determined by the resource increase / decrease target calculation unit;
A subsystem control node comprising:
A service providing system comprising:

Images