A Scheduling Mechanism Considering Simultaneous Running of Grid Tasks and Local Tasks in the Computational Grid

Task scheduling under a grid environment is an important research area, on which there are now many researches. Either in the meta-task scheduling problems or DAG (direct acyclic graph) scheduling problems, it is usually assumed that tasks or sub-tasks are submitted to dedicated and idle hosts and that these tasks are processed in the FIFO (first in first out) order. However, this is not practical in a grid, in which a host may be shared between grid users and its owner and execute grid tasks mixed with its own local tasks simultaneously. EBGSM (estimation based grid scheduling mechanism) is proposed, which allows for the simultaneous processing of grid tasks and local tasks. In EBGSM we use history information about the execution of tasks to predict the performance of non-dedicated hosts. We also applied EBGSM to MCT (minimum completion time) and min-min algorithms and obtain a smaller makespan, proving EBGSM practical for grid task scheduling.