ADOM: An Adaptive Objective Migration Strategy for Grid Platform

Object migration is the movement of objects from one machine to another during execution. It can be used to enhance the efficiency and the reliability of grid systems, such as to balance load distribution, to enable fault resilience, to improve system administration, and to minimize communication overhead. Most existing schemes apply fixed object migration strategies, which are unadaptable to changing requirements of applications. In this paper,we address the issue of object migration for large scale grid system with multiple object levels.First, we devise a probabilistic object tree model and formulate the object migration problem as an optimization problem. Then we proposed an adaptive object migration algorithm called ADOM to solve the problem. the ADOM algorithm applies the breadth first search scheme to traversal the object tree and migrates object adaptively according to their access probability. Finally we evaluate the performance of different object migration algorithms in our grid platform, which show that the ADOM algorithm outperforms other algorithms under large object tree size.