Research On PE Resource Control Policy In Heterogeneous Environment

One vital method of enhancing the performance of a distributed system is to improve its concurrency capability with resource control as a core. The resource can be refered to any hardwares and softwares only depending on weather they are necessary for the others. This paper limits the resource definition as the Processing Entity for the deeply dicussing.Central resource control makes the foucus on job scheduling which is the major branch in resource control. This paper studied on how to improve the central resource control polices.Central resource control polices have got a wonderful result for the local area networks. However, with the development of the networks, the running environment of the distributed systems has expanded from local area network to open, large scaled networks, such as p2p network. Central resource control policies have to face a huge gap between increasing job numbers and the computing bottleneck. Decentralized resource control policy inavoidably emerged to solve this problem. This paper also studied the decentralized resource control policies.The research work and creations of this paper are including as follows:1. Constructed a mathematical model for heterogeneous networks and PE resources with an instance BWRAP.There is an absent of an integrated mathematical model for the resource control policies which can effectively describe the relationship between jobs, resources, and environment. This paper gave a three step constructing method to uniform it. As an instance, a centralized resource control policy BWRAP is been given.2. Construced a centralized resource control policy—MMP which can ganrantee the Qos limits under Master/Slave architecture.MMP solved the discrete tasks' resource allocation under Master/Slave architecture. Because we construct its cost function as linear programming problem, the solving algorithm is not necessary as other central resource control policies.3. Constructed a centralized resource control policy-EDCP which can effectively checking the duplication process in DAG resource allocation.EDCP can improve the performance of centralized resource control policies based on duplication technologies by checking the necessity of its duplication.4. Constructed a decentralized resource control policy -CDRAP which also can treat DAG tasks. CDRAP using message broker+subscribe/Issue platform to descentralized allocating tasks' resources and can achieve a fairly balance result.5. Constructd a dynamic resource control model based on distributed cognition theory and analyzed 4 policies described above.The dynamic resource control model based on distributed cognition theory can pop out different characters of a same environment which can effect the executing of a given resource control policy.