Towards an opportunistic grid scheduling infrastructure based on tuple spaces

摘要

One main issue associated with the efficient and effective use of heterogeneous resources in a grid system is the scheduling. Scheduling in a grid system involves a number of challenging issues mainly due to the dynamic nature of the grid. Schedulers on traditional grid infrastructures rely on an information service that provides information about resources capacities and availability. However, in an asynchronous distributed system like a grid providing up-to-date information about resources is difficult. Current scheduling algorithms make scheduling decisions without fully accurate information about resources which can lead to inefficient schedules. This paper proposes a new scheduling infrastructure for grids where resources select tasks they execute, instead of the traditional approach where schedulers finding resources for the tasks. The new proposed approach allows, at any time, to make scheduling decisions with up-to-date/accurate information. Moreover, our infrastructure provides mechanisms to provide a fault tolerant scheduling. The proposed infrastructure is mainly based on the tuple space coordination model. In our evaluation study, a number of experiments with various simulation setting demonstrated the practicability of proposed infrastructure.