No abstract

Computer simulation is an important tool for studying epidemic dynamics. Owing to the scales and runtime speed requirements, the simulations of large-scale pandemic diseases such as severe acute respiratory syndrome (SARS) and H1N1 influenza usually require high-performance parallel simulation or computation. Previous works on the parallel simulations of large-scale epidemic were implemented on traditional general purpose CPU-based platforms or clusters. As more and more high-performance computation clusters are being built with so-called general-purpose graphics cards, this paper presents the implementation and optimization techniques for social contact network-based large-scale epidemic simulation on GPU clusters. Compared with previous works, this paper focuses on (1) how to efficiently implement the contact network-based parallel epidemic simulation on GPUs, and (2) how to hide communication latencies between processing nodes to improve scalability. Our proposed techniques are implemented and tested on a commodity cluster whose processing nodes are equipped with GPGPUs. The experimental results show that, for the simulation of 20 million individuals and 1.2 billion host contacts on 80 nodes, the execution on GPUs can achieve 7.4×−11.7× speedup over the execution on CPUs.

Compared with single-issue general purpose processors (GPPs), extensible multi-issue/VLIW processors can exploit instructionlevel parallelism, which are more suitable for computation intensive tasks. Moreover, they offer the ability of customizing computation accelerators for an application domain. In this paper, we present an automated methodology that customizes computation accelerators for the multi-issue/VLIW extensible processors, where several techniques are also proposed to optimize the design of an accelerator.

No abstract

No abstract