Applying internet random early detection strategies to scheduling in grid environments

Resource Allocation in Grid environments to date is generally carried out under the assumption that there is one primary scheduling system scheduling jobs. However, as environments tend towards larger open "utility" Grids it becomes increasingly likely that deployments will involve multiple independent schedulers allocating jobs over the same resources. In this paper we show that, if using current standard scheduling approaches, such multi-scheduler environments may well be prone to serious oscillation problems in resource allocation similar to those commonly found in IP network traffic. Further we demonstrate how common techniques from IP networks - in particularly approaches based on Random Early Detection (RED) buffer management and its subsequent extensions / variations - may provide an effective way to damp or eliminate such oscillations. The paper describes the analogy between multi-scheduler Grid resource allocation and IP network routing and explores the impact of oscillation and RED methods by simulation.