An investigation on runtime task scheduling for parallel raytracing on a heterogeneous distributed computing system

A case study is presented which is aimed at investigating the runtime task sizing problems for a heterogeneous Parallel Distributed Processing (PDP) system. The PDP is based on a client and servers scheme. In this paper the variants of workload partitioning and tasks scheduling strategies, in conjunction with different task granularities, are evaluated with their runtime behavior. The results indicate that the combination of roughly sub-task size estimation technique plus adaptive sub-task sizing strategy along with runtime machine performance probing estimation approach cope with both the unbalanced workload charactestics of the parallel raytracing application and the different computational capabilities of the machines in a workstations' cluster environment. Our proposed scheme cuts the efforts required to obtain the grained sub-task size by performing a number of experiments for specified PDP configuration. The results obtained front the proposed scheme are nearly equal to the best results. The performance of the PDP system is 2.5 times better as compared to the fastest machine in the network environment but still have 12% (at average) performance difference from the expected ideal.