Adaptive execution and performance tuning of parallel jobs in computational desktop grid using GridGain

Abstract

Desktop grids, which use the idle cycles of desktop PCs of small enterprises and institutions, usually are equipped with hundreds or thousands of desktops, mainly used for office tasks. Such grids vary markedly from conventional grids (multi-site clusters) in terms of their dynamic nature. This calls for the need of new scheduling algorithms, tailor-made for such systems. Since the nature of such desktop grids attract highly parallel algorithms, designing efficient scheduling algorithms of parallel jobs on desktop grids become worthwhile research issue. In this paper an adaptive scheduling mechanism for desktop grids has been presented, where tasks are initially assigned to different available grid resources based on a resource selection algorithm. The resource selection algorithm, which works in an online mode, selects resources based on given set of parameters and previous execution log in an adaptive (flexible) manner. Execution log helps to decide in choosing the most relevant computational parameter depending upon the tasks that are assigned to grid resources. Later, during task execution, if performance of any task degrades, then adaptive algorithm assists the scheduling policy to adapt the system to achieve the high throughput by rescheduling to either the server node or to the best available local node at present. This method can also deal with the unpredictable execution conditions commonly encountered on desktop grids. In this paper, GridGain 2.0 has been used for setting up desktop grid test bed and the performance of the aforesaid adaptive scheduling algorithm has been presented.