Methodology for Efficient Execution of SPMD Applications on Multicore Environments
CCGRID '10 Proceedings of the 2010 10th IEEE/ACM International Conference on Cluster, Cloud and Grid Computing
Efficient parallel selective separable-kernel convolution on heterogeneous processors
Proceedings of the Second International Forum on Next-Generation Multicore/Manycore Technologies
| Hi-index | 0.00 |
The significance of load-balancing is rising with the increasing number of processing cores per chip. A fast load-balancer is sought to exploit fine grain parallelism possible with multicore processors. This paper focuses on load-balancing image processing applications where the amount of processing varies per pixel; such application domain includes high dynamic range imaging and fractal related applications. This paper formulates load-balancing, for such applications, as a curve fitting problem for statistically sampled work across the image. Such formulation allows for analytically determining an optimal load-balance partitioning, resulting in accurate, low-overhead load-balancing. Experiments using the proposed algorithm shows a remarkable performance of achieving around 90% of 'perfect' image partitioning for a complex fractal application.