The main caveat of this approach is that by mapping jobs across cluster boundaries, inter-cluster network resources are also consumed. If the inter-cluster network links become too saturated with traffic, any co-allocated jobs may experience degraded runtime performance due to the communication bottleneck present in the network infrastructure. This degradation in runtime performance can potentially offset the benefit of performing job co-allocation in the first place. More precisely, the increase in job runtime due to link saturation can rapidly outweigh the decrease in queue waiting time, thus resulting in poorer overall system performance.

Multi-cluster schedulers must make use of all available information pertaining to job communication structure as well as network topology and utilization in order to improve job throughput while mitigating any negative impact to job runtime performance due to network congestion. Additionally, these schedulers must make reasonable co-allocation decisions in the absence of specific job and network information, as this information is not always available.

In this research, we have developed a bandwidth-centric job communication model that captures the interaction and impact of simultaneously co-allocating jobs across multiple clusters. We compare our dynamic model with previous research that utilizes a fixed execution time penalty for co-allocated jobs. We explore the interaction of simultaneously co-allocated jobs and the contention they often create in the network infrastructure of a dedicated computational multi-cluster.

We have also developed several bandwidth-aware co-allocating meta-schedulers. These schedulers take inter-cluster network utilization into account as a means by which to mitigate degraded job run-time performance. By making use of a bandwidth-centric parallel job communication model we are able to evaluate the performance of multi-cluster scheduling algorithms that focus not only on node resource allocation, but also on shared inter-cluster network bandwidth.

BeoSim is a discrete-event simulator that has been implemented for the purpose of studying multi-site parallel job scheduling algorithms in the context of a multicluster computational grid.