JP5551967B2 - Cluster system, cluster system scale-out method, resource manager device, server device - Google Patents

Description

  The present invention relates to a highly available cluster system that provides communication services, a cluster system scale-out method, a resource manager device, and a server device.

  In recent years, service diversification has been remarkable on the Internet. In addition, since it is difficult to clearly grasp the number of users in advance, the new service starts service provision (small start) using a small server first, and the server increases as the number of users increases. It is effective in terms of cost to increase the power.

  Also, in recent years, as a means of augmenting servers, instead of replacing hardware with more powerful ones, adding servers that perform the same processing (scaling out) and performing processing in clusters in which multiple servers operate in parallel Is becoming common.

  However, in order to quickly scale out when the number of users increases, it is necessary to always keep a standby server that is not operating in standby. It is costly for each service provider to prepare such a standby server that is not always used.

  Thus, in recent years, a form called IaaS (Infrastructure as a Service) or the like in which a server management company such as a data center collectively manages servers and hosts service providers has become common (Non-patent Document 1). ). In IaaS, not only hardware maintenance of a server is unnecessary for each service provider, but also a reduction in server facility maintenance costs can be expected by sharing a standby server among a plurality of service providers.

  On the other hand, in recent years, IP of communication networks has been promoted for the purpose of reducing equipment maintenance costs, and it has become common to implement communication networks and communication services using general-purpose servers instead of dedicated hardware like conventional exchanges. ing. In addition, as communication services are diversified, there is a demand to reduce equipment costs by starting small new services and scaling out the number of servers according to the expansion of users. In response to such a request, it is possible to apply a configuration in which each service is started in a small manner, and when a user of a certain service is expanded, a necessary server is acquired from a resource pool in which surplus standby resources are aggregated and scaled out. .

Amazon Elastic Compute Cloud (Amazon EC2), http://aws.amazon.com/ec2/.

  In a configuration in which a server is acquired from a shared resource pool in which surplus resources are aggregated at the time of scale-out of individual services as described above, the server held in the resource pool is shared, and what software Since it is not known in advance whether to take the configuration, it is common to have no software. Therefore, after acquiring the server at scale-out, transfer the software image containing the software configuration corresponding to each service from the repository that manages the image to the server, make the necessary settings, and incorporate it into the cluster There is a need.

  Since the software image includes all of the operating system, various middleware, data, and the like, the software image has a large capacity ranging from several to several tens of GB, and transfer requires a certain amount of time. When multiple services are scaled out at the same time, the repository that manages the image becomes a bottleneck, and the scale-out time increases. Therefore, non-stop and responsiveness required for highly public communication services as infrastructure Can't meet.

  In view of the above-described problems, an object of the present invention is to provide a cluster system, a cluster system scale-out method, a resource manager device, and a server device that can reduce the load of software image transfer during scale-out. To do.

In order to solve the above-described problems, a cluster system according to the present invention is a cluster system including a plurality of servers that provide services and a resource manager that manages the plurality of servers. The resource manager performs a predetermined service. If there is a request for scale-out to add a server, a server in a standby state is selected as a server to be scaled out from among a plurality of servers, and a predetermined service is already activated in the plurality of servers. Means for instructing a plurality of servers to transfer software image fragments of a predetermined service, and the plurality of servers instructed to transfer software scale out software image fragments of a predetermined service A means for transferring to a server elected as the server, Server elected as a server for performing is characterized in that it comprises means for receiving and storing software image for a given service that has been transferred from a plurality of servers are instructed to transfer the software.

The cluster system scale-out method according to the present invention adds a server for performing a predetermined service in the cluster system scale-out method including a plurality of servers providing services and a resource manager managing the plurality of servers. When a scale-out request is made, the resource manager selects a server in a standby state from a plurality of servers as a server to be scaled out, and a predetermined service is already operating in the plurality of servers. Instructing a plurality of servers to transfer software image fragments of a predetermined service, and a plurality of servers instructed to transfer software transfer software image fragments of a predetermined service to a server that performs scale-out And the server that performs the scale-out, Characterized in that it comprises a step of receiving and storing software image for a given service transferred from software plurality of servers that are instructed to transfer.

A resource manager device according to the present invention manages a software image necessary for service execution in a resource manager device for configuring a cluster system from a plurality of servers that provide services and a resource manager that manages the plurality of servers. If there is a scale-out request for adding a means, a means for monitoring the operating status of a plurality of servers, and a server for performing a predetermined service, the server in a standby state is scaled out from the plurality of servers. Means for selecting a server and requesting a plurality of servers already operating a predetermined service among the plurality of servers to transfer a piece of software image of the predetermined service to a server performing scale-out It is characterized by providing.

A server device according to the present invention includes a means for transferring and acquiring software image fragments in a server device for configuring a cluster system from a plurality of servers that provide services and a resource manager that manages the plurality of servers. The software image storage means and the individual setting information storage means, and when the other server is set as a scale-out server, the software stored in the software image storage means according to the instruction of the resource manager transfer the pieces of image server for scale-out, when the server is selected as the server for scale-out, already acquired the software images transferred from a plurality of other servers are operating a predetermined service As a software image storage means. Characterized in that it.

  According to the present invention, at the time of scale-out, a software image used by an active server that is already in operation is transferred to a new scale-out server, and a unique setting of the new scale-out server is performed. Is called. In addition, a software image is partially transferred from a plurality of servers constituting the cluster to a new scale-out server. For this reason, the transfer load is not concentrated on a specific active server, and it is possible to scale out at high speed.

1 is a block diagram illustrating an overall configuration of a cluster system according to a first embodiment of the present invention. It is explanatory drawing of a structure of the management information table of a software image. It is explanatory drawing of a system when the service has not started in the cluster system which concerns on the 1st Embodiment of this invention. It is explanatory drawing of the server state of the server state monitoring function part when the service has not started in the cluster system which concerns on the 1st Embodiment of this invention. It is a sequence diagram used for description when starting service "A" in the cluster system which concerns on the 1st Embodiment of this invention. It is explanatory drawing of the server state of the server state monitoring function part when service "A" is started in the cluster system which concerns on the 1st Embodiment of this invention. It is a sequence diagram used for description when starting service "B" in the cluster system which concerns on the 1st Embodiment of this invention. It is explanatory drawing of a system when service "B" is started in the cluster system which concerns on the 1st Embodiment of this invention. It is explanatory drawing of the server state of a server state monitoring function part when service "B" is started in the cluster system which concerns on the 1st Embodiment of this invention. FIG. 7 is a sequence diagram used for explanation when a service “B” is scaled out in the cluster system according to the first embodiment of the present invention. It is explanatory drawing of a system when the service "B" is scaled out in the cluster system which concerns on the 1st Embodiment of this invention. It is explanatory drawing of the server state which the server state monitoring function part has when the service "B" is scaled out in the cluster system according to the first embodiment of the present invention.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing the overall configuration of the cluster system according to the first embodiment of the present invention.

  In FIG. 1, the cluster system 102 according to the first embodiment of the present invention is connected to the maintenance person terminal 101 via the maintenance network 150 and is connected to the user terminal 140 via the service network 160.

  The service provider executes a maintenance command for the cluster system 102 using the maintenance person terminal 101. Maintenance commands include building a server cluster to provide services and augmenting the cluster by scaling out.

  A user uses a service provided on the cluster system 102 via the service network 160 using the user terminal 140. The cluster system 102 includes a resource pool 120 in which calculation resources are aggregated, and a resource manager 103 that is a function unit for managing resources, and both are connected by a management network 110. The resource pool 120 is configured by a plurality of servers 130 that are computational resources.

  Next, the configuration function unit of the resource manager 103 will be described. The resource manager 103 includes a management interface 201, a software image management function unit 202, a server state management function unit 203, a server activation processing function unit 204, and a server state monitoring function unit 205.

  The management interface 201 performs an interface with the maintenance person terminal 101, receives a maintenance command from the maintenance person terminal 101, and executes appropriate processing in cooperation with other function units. The software image is a file including a software configuration necessary for providing a service, and is transferred to the server 130 when the service is provided.

  The software image management function unit 202 manages the type and storage location of a software image necessary for executing a predetermined service. FIG. 2 shows an example of a software image management information table held by the software image management function unit 202. One software image exists for each service. In the example of FIG. 2, the service “A” starts the service using the image “A”, and the service “B” uses the image “B”. Further, the storage location of the images “A” and “B” is “/ path / to / imageDir” on the file system of the resource manager 103. In this embodiment, the storage location is assumed to be a disk or the like on the resource manager 103. However, the actual storage location may be an external storage or the like as long as it can be accessed from the resource manager 103.

  The server state management function unit 203 manages state information of the server 130 such as which service the server 130 that is a resource in the resource pool 120 is currently used for.

  When there is a server activation request, the server activation processing function unit 204 refers to the information of the server status monitoring function unit 205 to determine which server 130 in the resource pool 120 is to be used and The software image management function unit 202 has a function of transferring and starting a software image managed by the software image management function unit 202. In addition, when there is a scale-out request, the server activation processing function unit 204 refers to the information of the server state monitoring function unit 205 and determines which server 130 in the resource pool 120 is used for the scale-out. The server 130 having the same service has a function of requesting the server 130 used for scale-out to transfer a piece of software image and starting it.

  The server state monitoring function unit 205 has a function of constantly monitoring the state of the server 130 in the resource pool 120 and updating the server state management function unit 203. 4, 6, 9, and 12 show examples of server states that the server state monitoring function unit 205 has. In FIG. 4, the states of the six servers 130 with the server IDs “SV1” to “SV6” are all on standby. In FIG. 6, the server 130 with the server IDs “SV1” and “SV2” is executing the service “A”, and the other servers are waiting. In FIG. 9, the server 130 with server IDs “SV1” and “SV2” is executing “service“ A ”, and the server 130 with server IDs“ SV4 ”and“ SV6 ”is executing service“ B ”. And other servers are waiting. In FIG. 12, a server 130 with server IDs “SV1” and “SV2” is executing “service“ A ”, and a server 130 with server IDs“ SV4 ”,“ SV5 ”, and“ SV6 ”is service“ B ”. ”And other servers are waiting.

  Next, the configuration function unit of the server 130 in the resource pool 120 will be described. The server 130 includes an activation agent 301, a software image storage unit 302, an individual setting / data storage unit 303, and a calculation processing function unit 304.

  The activation agent 301 performs communication between the resource manager 103 and the server 130. Specifically, it receives a start instruction from the server start processing function unit 204 and acquires a software image. Further, in response to the status inquiry from the server status monitoring function unit 205, the service type being executed on the server 130 or being in standby is notified. When a software image transfer instruction is issued from the server activation processing function unit 204, the software image fragment is transferred to another server 130.

  The software image storage unit 302 is a functional unit that stores the software image acquired by the activation agent 301. The software image exists for each service, but the setting information necessary for starting the server 130 and executing the service may differ for each individual server 130. In this system, the setting information for each server 130 is stored in the individual setting / data storage unit 303 instead of rewriting and storing the acquired software image. Individual data such as log information that is output differently for each server 130 is also arranged in the individual setting / data storage unit 303.

  The calculation processing function unit 304 is a calculation device such as a CPU or main memory of the normal general-purpose server 130, and boots the system using the software image storage unit 302 and the individual setting / data storage unit as disks to perform service providing processing. .

  Next, the operation of the cluster system 102 according to the first embodiment of this invention will be described. FIG. 3 shows a state in which no service is activated in the cluster system 102 having the resource pool 120 composed of six servers 130 (server IDs “SV1” to “SV6”). In the cluster system 102, when no service is activated, the server status information held by the server status monitoring function unit 205 includes six servers 130 having server IDs “SV1” to “SV6” as shown in FIG. All of the states are waiting. In the following, processing when transitioning from this state to a state in which service “A” and service “B” are being executed will be described.

  FIG. 5 is a sequence diagram showing processing when the six servers 130 (server IDs “SV1” to “SV6”) transition from the standby state to the state where the service “A” is being executed.

  When the resource manager 103 receives, from the maintenance person terminal 101, an activation request by the two servers of the service “A” via the management interface as a maintenance command (step S101), the server activation processing function unit 204 causes the server state monitoring function unit 205 to The two servers 130 that are in a standby state are selected (step S102). Here, it is assumed that servers 130 having server IDs “SV1” and “SV2” are selected.

  When the servers 130 with the server IDs “SV1” and “SV2” are selected, the server activation processing function unit 204 of the resource manager 103 uses the software image necessary for executing the service “A” from the software image management function unit 202. An image “A” is specified, and this image “A” is transferred to the activation agent 301 of the selected server 130 (server ID = SV1, SV2) (steps S103a and S103b). The activation agent 301 on the selected server 130 (server ID = SV1, SV2) stores the received image “A” in the software image storage unit 302 (steps S104a and S104b).

  The server startup processing function unit 204 of the resource manager 103 changes the network settings such as the IP address, netmask, and DNS server IP address, and the middleware and OS parameters such as the buffer size and the number of simultaneous connections before starting the server 130. When the individual setting information that needs to be passed is passed to the activation agent 301 on the server 130 with the server IDs “SV1” and “SV2” (steps S105a and 105b), the server 130 with the server IDs “SV1” and “SV2”. The activation agent 301 above stores the individual setting information in the individual setting / data storage unit 303 (steps S106a and 106b).

  Thereafter, the activation agent 301 of the server 130 with the server IDs “SV1” and “SV2” instructs the calculation processing function unit 304 to boot using the software image and the individual setting information. The calculation processing function unit 304 operates as the server 130 that provides the post-boot service “A”.

  As described above, when the server 130 that provides the service “A” is set, the server status information of the server status monitoring function unit 205 includes six servers 130 (servers) in the resource pool 120 as illustrated in FIG. Among the servers having IDs “SV1” to “SV6”), the servers 130 with the server IDs “SV1” and “SV2” are executing the service “A”, and the other servers are on standby.

  Next, when the server 130 that provides the service “B” on the server IDs “SV4” and “SV6” is activated in response to the activation request maintenance command for the service “B” in the same procedure, FIG. In step S201, steps S206a and S206b are performed, and the state shown in FIG. 8 is obtained. As shown in FIG. 9, the server status information of the server status monitoring function unit 205 at this time is the server ID among the six servers 130 (server IDs “SV1” to “SV6”) in the resource pool 120. Servers “SV1” and “SV2” are currently executing service “A”, servers 130 whose server IDs are “SV4” and “SV6” are currently executing service “B”, and other servers are waiting. It becomes.

  Next, the procedure for scaling out service “B” from this state will be described. FIG. 10 is a sequence diagram illustrating processing when the service “B” is scaled out.

  When the resource manager 103 receives the maintenance command for the scale-out request for the service “B” via the management interface (step S301), the server activation processing function unit 204 of the resource manager 103 refers to the information of the server state management function unit 203. Then, one waiting server 130 is selected (step S302). As a selection method, a simple method such as selecting one server 130 in a waiting state at random, selecting the first one found while checking the table entries in order, The activation processing function unit 204 knows the actual network topology and the physical location of the server 130 in the resource pool 120, and can determine that the communication delay is the smallest for the active server 130 based on the information ( For example, a method of selecting the server 130 (connected to the same hub or directly connected by a network cable) is also conceivable.

  Here, it is assumed that the server 130 with the server ID “SV5” is selected as the server to be scaled out from the waiting servers 130. In this case, the server 130 with the server ID “SV5” needs to transfer the software image of the image “B”. In a normal server activation request, the server activation processing function unit 204 of the resource manager 103 directly transfers the software image necessary for executing the service “B” from the software image management function unit 202 to the server 130 with the server ID “SV5”. However, at the time of scale-out, the server activation processing function unit 204 does not directly transfer the software image, and the server providing the service “B” requests transfer of a fragment of the image “B”.

  That is, in this case, the server activation processing function unit 204 of the resource manager 103 sends the image “B” to the activation agent 301 for the server 130 (server ID = SV4, SV6) that already has the image “B”. ") Is transferred (steps S303a and 303b).

  The server 130 (server ID = SV4, SV6) holding the image “B” transfers the instructed fragment of the image “B” to the server 130 whose server ID is “SV5” (steps S304a and 304b). At this time, the amount of fragments transferred by each server 130 can be divided equally and a method of weighting can be considered. In the case of equally transferring, assuming that the capacity of the image “B” is 1 GB, the server 130 with the server ID “SV4” stores data up to 500 MB, which is half the head of the image “B”, and the server ID is “ The server 130 with “SV6” transfers 500 MB from the half to the end of the image “B” to the server 130 with the server ID “SV5”.

  When weighting is performed, a method of using static information of each server and a method of dynamically considering the processing load of each server can be considered. The former is based on the premise that the server activation processing function unit 204 knows in advance the processing load and idle resources (CPU time, network bandwidth, etc.) that each server is performing on average, and the latter is based on the activation agent 301. The server activation processing function unit 204 grasps the server processing load and availability resources at that time. Then, when the number of idle resources of the server 130 with the server ID “SV4” is, for example, three times that of the server 130 with the server ID “SV6”, and the capacity of the image “B” is 1 GB, the server ID is “SV4”. The server 130 with the server ID “SV6” transfers the remaining 250 MB to the server 130 with the server ID “SV5”.

  The activation agent 301 of the server 130 with the server ID “SV5” stores the received image “B” in the software image storage unit 302 (step S305). Then, the server activation processing function unit 204 of the resource manager 103 performs network settings such as an IP address, a netmask, an IP address such as a DNS server, middleware and OS parameters such as a buffer size and the number of simultaneous connections, and the like before the server 130 is activated. Is passed to the activation agent 301 on the server 130 whose server ID is “SV5” (step S306), the activation agent 301 on the server 130 whose server ID is “SV5” The individual setting information is stored in the setting / data storage unit 303 (step S307).

  Thereafter, the activation agent 301 of the server 130 with the server ID “SV5” instructs the calculation processing function unit 304 to boot using the software image and the individual setting information. The calculation processing function unit 304 operates as the server 130 that provides the post-boot service “B”.

  At the time of scale-out, processing as shown in steps S301 to S307 is performed, and a state as shown in FIG. 11 is obtained. The server status information of the server status monitoring function unit 205 at this time is, as shown in FIG. 12, the server ID of the six servers 130 (server IDs “SV1” to “SV6”) in the resource pool 120. Servers “SV1” and “SV2” are executing service “A”, and servers 130 whose server IDs are “SV4”, “SV5”, and “SV6” are executing service “B”, and other servers Will be waiting.

  As described above, it is possible to simultaneously scale out a plurality of services at high speed by not applying a software image transfer load to the resource manager 103 during scale-out.

  The present invention is not limited to the above-described embodiments, and various modifications and applications can be made without departing from the gist of the present invention.

101: maintenance person terminal 102: cluster system 103: resource manager 110: management network 120: resource pool 130: server 140: user terminal 150: maintenance network 160: service network 201: management interface 202: software image management function unit 203: Server state management function unit 204: Server activation processing function unit 205: Server state monitoring function unit 301: Activation agent 302: Software image storage unit 303: Data storage unit 304: Calculation processing function unit

Claims (9)

In a cluster system comprising a plurality of servers that provide services and a resource manager that manages the plurality of servers,
When there is a request for scale-out to add a server for performing a predetermined service, the resource manager selects a server in a standby state from among the plurality of servers as a server to perform scale-out, and Means for instructing a plurality of servers already operating the predetermined service in the server to transfer a piece of the software image of the predetermined service;
The plurality of servers instructed to transfer the software includes means for transferring a piece of software image of the predetermined service to a server selected as the server that performs the scale-out,
The server selected as the server to perform the scale-out has means for receiving and storing a software image of a predetermined service transferred from a plurality of servers instructed to transfer the software. system.   There are a plurality of servers instructed to transfer the software, and each of the plurality of servers instructed to transfer the software scales out a piece of the software image of the predetermined service with an equal amount of transfer. The cluster system according to claim 1, wherein the cluster system is transferred to a server.   There are a plurality of servers instructed to transfer the software, and each of the plurality of servers instructed to transfer the software performs a scale-out of a piece of the software image of the predetermined service with a weighted transfer amount. The cluster system according to claim 1, wherein the data is transferred to the cluster system.   The cluster system according to claim 3, wherein the weighting is performed according to a processing load of the server and a free resource.   5. The cluster system according to claim 4, wherein the processing load of the server and information on available resources are statically managed by the resource manager based on an average processing load and available resources of each server.   The cluster system according to claim 4, wherein the processing load of the server and information on holidays are managed dynamically by the resource manager by acquiring information from each server. In a cluster system scale-out method comprising a plurality of servers for providing services and a resource manager for managing the plurality of servers,
When there is a scale-out request to add a server for performing a predetermined service, the resource manager selects a server in a standby state from the plurality of servers as a server to perform scale-out, and the plurality of the plurality of servers. Instructing a plurality of servers already operating the predetermined service in the server to transfer a piece of the software image of the predetermined service;
A plurality of servers instructed to transfer the software, transferring a piece of software image of the predetermined service to the server performing the scale-out;
A scale-out method comprising: a server performing the scale-out receiving and storing a software image of a predetermined service transferred from a plurality of servers instructed to transfer the software. In a resource manager device for configuring a cluster system from a plurality of servers that provide services and a resource manager that manages the plurality of servers,
A means of managing software images required for service execution;
Means for monitoring operating states of the plurality of servers;
When there is a request for scale-out to add a server for performing a predetermined service, a server in a standby state is selected as a server to perform scale-out from among the plurality of servers, and the server among the plurality of servers is selected. Means for requesting a plurality of servers already operating a predetermined service to transfer a piece of a software image of the predetermined service to a server that performs the scale-out. apparatus. In a server device for configuring a cluster system from a plurality of servers that provide services and a resource manager that manages the plurality of servers,
Means for transferring and obtaining pieces of software images;
Software image storage means;
Storage means for individual setting information,
When another server is set as a server to be scaled out, according to an instruction from the resource manager, a piece of software image stored in the software image storage means is transferred to the server to be scaled out,
When the server is selected as a server to be scaled out, a software image transferred from a plurality of other servers already operating a predetermined service is acquired and stored in the software image storage means. A server device as a feature.

Images