| 摘要 |
PURPOSE: A method and an apparatus for resource management in grid computing systems are provided to optimize quality of service by defining deadlines of tasks and users' budget availability for grid services as SLA(Service Level Agreements). CONSTITUTION: SLA information and resource state information respectively obtained by a service level management unit(100) and a resource state management unit(200) are provided to an expected completion time calculation unit(311). The expected completion time calculation unit recognizes a size of the task and real-time states of resources in the grid by using the SLA information and the resource state information, and calculates for each resource the expected completion time of the task based on a computing speed thereof. An expected profit calculation unit(312) calculates the expected profit for each resource by calculating a total cost for use in executing the task and subtracting the total cost from a user's budget to be paid for completing the task, wherein the total cost is sum of a penalty for missing the deadline of the task and an expected cost calculated by using the expected completion time and a billing policy. An available resource cluster creation unit(313) creates the available resource cluster by excluding, based on the expected completion time and the expected profit, resources which do not satisfy the user's requirements from the entire resources in the grid. A task division unit(314) divides the task into multiple partial tasks, if the available resource cluster cannot be created by the available resource cluster creation unit. A task processing policy adjustment unit(315) adjusts, when the SLA information cannot be satisfied by the available resource cluster, the SLA information via a negotiation process on service quality with the user. A task processing policy creation unit(316) creates a policy for processing the task, based on the expected completion time, the expected profit and the available resource cluster, and when necessary, further based on the task division result by the task division unit and the negotiation result by the task processing policy adjustment unit. A task management unit(320) includes a task processing policy storage(321), a task processing policy execution unit(322), a task execution unit(323) and a task monitoring unit(324).
|
