JP3977298B2 - Grid computing system - Google Patents

Description

  The present invention relates to a grid computing system that dynamically combines a plurality of computing resources scattered on a network to form a virtual machine, and a computing resource collection method in the grid computing system.

  In grid computing, in order to determine which computing resource to request a job (calculation processing, etc.), information on the state of the computing resource (eg, CPU usage rate) is managed, and the work is performed according to the information. Make a request. Conventionally, status information of all computing resources is managed by a directory prepared in a specific management server.

  For example, in the computing resource collection function (MDS: Monitoring and Discoverry Service (Information Service)) in Globus Tool Kit (registered trademark) recognized as a de facto standard as middleware constituting a grid computing system, it is shown in FIG. The directory inquiry method is used.

  In FIG. 3, each of the computing resources 100 to 103 periodically reports information on the state of its own computing resources (for example, CPU usage rate) to the directory 110 prepared in the central management server. The user 120 who intends to use the computing resources 101 to 103 refers to the directory 101 and searches for available computing resources (discovery). When an available computing resource is found, the individual computing resource is accessed (lookup) (see, for example, Non-Patent Document 1).

  When the user 120 collects information on the computing resources 101 to 103 by the conventional inquiry method to the directory described above, there are the following problems.

First, when the directory is abnormal, the entire system managed by the directory is damaged. Secondly, the status information of the computing resources described in the directory is not always the latest. Thirdly, when the directories are hierarchized, there is a problem that it is difficult to manage the entire directory.
Globus Toolkit 2.2 MDS Technology Brief Draft 4-January 30 2003, [online], January 30, 2003, [Search July 16, 2003], Internet <URL: http://www.globus.org/mds /mdstechnologybrief_draft4.pdf>

  As described above, in the grid computing system using the conventional directory, it is necessary to update the state information of the computing resource as appropriate. However, if the real-time property of the update is to be improved, communication for transmitting the information is performed. There is a problem that traffic increases. In addition, since the status information of each computing resource is managed by a single directory, if the device in this directory part fails, the entire system will not function.

  The present invention has been made to solve such a problem, and an object of the present invention is to be able to find an optimal computing resource in real time, to improve the fault tolerance by eliminating the need for a directory, An object of the present invention is to provide a grid computing system and a computing resource collection method in the grid computing system that can continue services without affecting the entire system even when a computing resource fails.

In the grid computing system according to the present invention, each of a plurality of computing resources retains its own state information, updates the state information, and a requesting computer that requests the computing resource to perform processing Means for transmitting requirement information related to the state of the resource to the plurality of computing resources, and the computing resource that has received the requirement information satisfies the requirements based on the requirement information and the state information. Comprises means for returning an affirmative response to the requesting computer, and means for requesting the computing resource for which the requesting computer has returned an affirmative response, wherein the computing resource comprises the status information. And the requirement information is a search condition based on a file name that satisfies a predetermined requirement.
As a result, state information (CPU usage rate, etc.) can be updated on demand for computational resources, and the computer that requests the processing can discover the optimal computational resources in real time. Performance can be improved. In addition, since the conventional directory is not required, dynamic information exchange and lookup procedures between the computing resource and the directory are reduced, and extra traffic is reduced. Furthermore, by eliminating the need for a directory, the failure point (point of failure) is reduced, fault tolerance is improved, and even when individual computing resources fail, the entire system is not affected. It becomes possible to continue the service.

Further, this makes it possible to apply the file search function of the pure P2P system to search for requirement information (for example, CPU usage rate).

In the grid computing system and the calculation resource collection method of the present invention, each calculation resource retains its own state information (for example, CPU usage rate) and updates the state information as appropriate. The requesting computer that requests processing sends requirement information (for example, CPU usage rate) regarding the state of the computing resource to each computing resource, and each computing resource that has received the requirement information has its own requirements. A response is returned to the requesting computer, and the requesting computer requests the processing resource to respond.
As a result, state information (CPU usage rate, etc.) can be updated on demand for computational resources, and the computer that requests the processing can discover the optimal computational resources in real time. Performance can be improved. Further, since the conventional directory is not necessary, dynamic information exchange and lookup procedures between the computing resource and the directory are reduced, leading to a reduction in extra traffic. Furthermore, by eliminating the need for a directory, the failure point (point of failure) is reduced, fault tolerance is improved, and even when individual computing resources fail, the entire system is not affected. It becomes possible to continue the service.

Further, in the grid computing system of the present invention, the calculation resource of the calculation resource holds the contents of the state information (for example, CPU usage rate) as a file name, and the request source computer uses the request information as the requirement information (search condition). Use the file name.
Thereby, the file search function of the pure type P2P system can be applied to search for requirement information (for example, CPU usage rate).

  In the grid computing system according to the present invention, state information of computing resources is mutually exchanged by P2P (Peer to Peer) type communication. Each computing resource has a file with a predetermined file name corresponding to its own state (CPU usage rate, etc.). Further, by searching for a file name using software for exchanging files with P2P (for example, Gnutella (registered trademark)), the requesting computer searches for a computing resource in a predetermined state, and the corresponding state Request work (processing) for the computing resource.

  FIG. 1 is a diagram for explaining a grid computing system according to the present invention, and shows an example configured using a pure P2P system such as Gnutella (registered trademark). The file name search in Gnutella (registered trademark) has a mechanism in which file name searches are performed in a chained manner. In the grid computing system of the present invention, this function is used for status information (for example, CPU) of computing resources. (Use rate) is extended so that it can be searched by chaining by file name.

  In FIG. 1, the computing resources 1 to 13 have participated in Gnutella Net (registered trademark) in advance, measured the CPU usage rate in the background, and named the file name corresponding to the value as a pure P2P system (Gnutella (registered trademark)). Trademark) etc.) is stored in a predetermined directory.

  The requesting computer 20 for requesting calculation processing first participates in Gnutella Net (registered trademark), and then assigns a computational resource having an empty CPU necessary for assigning a process (for example, a task) to the Gnutella Net (registered trademark). Search by broadcasting to). The computing resources 1 to 13 satisfying the request respond to the requesting computer 20, and the requesting computer 20 assigns the computing process to the most suitable computing resource.

  Further, in the grid computing system of the present invention, the file search function of the pure P2P system is applied to the search of the CPU usage rate, etc., so how to reflect the CPU usage rate in the file name is a point.

  For this reason, the computing resources 1 to 13 regularly measure the CPU load (usage rate) of the terminal itself in the background. Further, a file having a file name based on Table 1 shown in FIG. 2A is created from the obtained value, and stored in a directory designated by a pure P2P system (Gnutella (registered trademark) or the like).

  In the example shown in FIG. 2, the file name of the file to be created and the search key have the following relationship. For example, when the CPU usage rate is 90% to 100%, the CPU is almost free, so that it is possible to search only with the search key “J” used when searching for a calculation resource of “almost free” or more. Thus, the file name to be created includes a search key “J” and a file name “J” that does not include other search keys. Further, when the CPU usage rate is 80% or more and less than 90%, the CPU is at least 10% free. Therefore, the search keys “J” and “ The file name to be created includes the search keys “J” and “I” so that the search can be made only with the search key “I” used when searching for a calculation resource of “10% or more free”. The file name does not include the search key “IJ”. Further, when the CPU usage rate is 70% or more and less than 80%, the CPU is at least 20% free. Therefore, the search keys “J”, “ Search is possible only with the search key “I” used when searching for a calculation resource of “10% or more free” and the search key “H” used when searching for a calculation resource of “20% or more free”. Thus, the file name to be created is the file name “HIJ” that includes the search keys “J”, “I”, and “H” and does not include other search keys.

  A script for performing the processing described above is created and executed on each of the calculation resources 1 to 13. With this function, the CPU load state is periodically saved as a file name. By searching for the file using a pure P2P system such as Gnutella (registered trademark), the calculation considering the free state of the CPU is performed. Search for resources becomes possible.

  Further, when searching for the calculation resources 1 to 13 from the request source computer 20, a search key based on Table 2 shown in FIG. 2B is used. For example, in a computational resource in which the CPU is at least 80% idle, the file name to be created always includes “B”. Therefore, if “B” is searched from the request source computer 20 as a search key. A computing resource whose CPU state is in an idle state of 80% or more is discovered. Then, the request source computer 20 assigns a calculation processing process to the found calculation resource.

  As described above, in the grid computing system of the present invention, the computing resource collection function in the grid computing system is realized by using a pure P2P system. Compared with the directory-type computing resource collection function that uses a management server that is generally used in conventional grid computing systems, an information retrieval method based on a brokerless concept in which there is no uniform information provider is used. In the present invention, it is possible to construct a grid computing system that can cope with dynamic fluctuations in computing resources and has excellent fault tolerance. In addition, since this method eliminates the need for a directory, there is a merit that management becomes easy.

  In the embodiment of the present invention, an example in which the CPU usage rate is used as the state information of the calculation resource is shown. However, the present invention is not limited to this, and various other types such as a memory capacity, a calculation speed, and a communication speed are used. State information can also be used, and state information combining these can also be used.

Although the embodiment of the present invention has been described above, the grid computing system of the present invention is not limited to the above illustrated example, and various modifications can be made without departing from the gist of the present invention. Of course.
The computing resource and the requesting computer in the present application are not limited in name and form as long as they are hardware having a computing function and a network connection function such as a personal computer, a PDA, and a mobile phone. The network includes both wired and wireless connection methods. The computing resource and the requesting computer can be replaced by a job.

  The grid computing system and the computing resource collection method in the grid computing system of the present invention have the effects of improving the performance of the grid computing system and increasing the fault tolerance.

  Therefore, the present invention can be applied to a grid computing system, a calculation resource collection method in the grid computing system, and the like.

The figure for demonstrating the grid computing system by this invention. The figure which shows the example of the file name and search key which a calculation resource produces. The figure which shows the general resource management mechanism of the conventional grid computing system.

Explanation of symbols

1-13 Computing resource 20 Requester computer

Claims (1)

A plurality of computing resources each holding its own state information and updating the state information;
A requesting computer for requesting processing to the computing resource, means for transmitting requirement information regarding the state of the computing resource to the plurality of computing resources;
Means for returning an affirmative response to the requesting computer when the computing resource that has received the requirement information satisfies the requirement based on the requirement information and the state information;
Means for requesting the computing resource to which the requesting computer has returned an affirmative response;
The grid computing system, wherein the calculation resource holds a file whose file name is the content of the state information, and the requirement information is a search condition based on a file name that satisfies a predetermined requirement.

Images