Grid scheduling using 2-phase prediction (2PP) of CPU power

Divisible workloads are that kind of workloads that can be partitioned by the scheduler into arbitrary 'chunks'. The problem of scheduling divisible loads has been defined for a long time, however, handful solutions have been proposed. Furthermore, almost all proposed approaches attempt to perform scheduling in a dedicated environment (i.e., for processing local tasks only) such as a LAN, whereas scheduling in non-dedicated environments (i.e., for processing local and external tasks) such as Grids remains an open problem. In Grids, the incessant variation of workstation's power is the chief difficulty in planning how to split and distribute workloads to these workstations. This paper presents a new strategy, called 2-Phase Prediction (2PP) for CPU power. By integrating this strategy and the UMR algorithm, a static scheduling algorithm that is designed for dedicated environments, we develop a new dynamic scheduling algorithm suitable for non-dedicated environment. Our experimental results show that our algorithm is superior to the UMR as the former is able to adapt to the dynamicity of Grid workers.