Adaptive load sharing in homogeneous distributed systems
IEEE Transactions on Software Engineering
| Hi-index | 0.00 |
Computing resource federation among collaborators is necessary for smooth promotion of collaborations. However, this is difficult for the collaborators who are using different type grid infrastructures, because of incompatibilities of the grid middleware. Therefore an inter grid job submission specification named HPC Basic Profile (HPCBP) has been defined by the Open Grid Forum (OGF) and many grid projects have implemented it. However, there still are many problems in the grid interoperation using the HPCBP. One of them is the workload disruption problem. The interoperation architecture, which is popular in the implementation of many prototypes, has a race condition between detection of the job submission from another grid and resource allocation for a submitted job from local client. This race condition disrupts the workload balance among the computing resources, and increases number of waiting jobs. In this paper, we explain and analyze the workload problem by an experiment and a simulation, and propose an architecture which can solve the problem, and show the effectiveness of the architecture by a simulation.