| Abstract: | Desktop Grids have the potential to provide an effective, low-cost solution
to the computing needs of a variety of distributed applications
consisting in a set of independent tasks. However, such a potential can be
exploited in practice only if suitable scheduling strategies are employed.
In this paper we propose a set of emph{knowledge-free} scheduling algorithms
(that is, they do not require any information concerning either resource status
or application characteristics) that
are able to effectively schedule a set of competing applications
simultaneously submitted to a Desktop Grid, unlike previous solutions that
are able to deal only with a single application at a time.
We study, via simulation, the performance of these strategies, as well as their ability
of efficiently using the available resources, for a wide
range of Desktop Grid configurations (ranging from Volunteer Computing
systems to Enterprise Desktop Grids) and application workloads.
Our results show that the scheduling algorithms we propose outperform
naive alternative like FCFS (the emph{de facto} standard for Desktop Grids)
both in terms of application performance and efficient resource usage.
Furthermore, these results enable us to identify which one of the
proposed algorithms should be used for a given combination of
Desktop Grid configuration and application workload. |
