Generalizing Gustafson's Law For Heterogeneous Processors

• André Cesta Eldorado Research Institute
• Geraldo Silva Eldorado Research Institute

Abstract

In this article we show how Gustafson's Law can be generalized and extended from a function of the number of processors 'p' to a function of intrinsic microprocessor variables such as: clock speed 'c' and hardware threads 't'. This allows that the performance or benchmark results following Gustafson's Law be modeled for heterogeneous processors. Some possible applications of this theoretical work include all previous applications from Gustafson's Law and some others: hardware upgrade or migration scenario analysis; optimal hardware selection for particular workloads; sizing or recommendations for heterogeneous machines.

References

Amdahl, G. M. (1967) "Validity of the Single-Processor Approach To Achieving Large Scale Computing Capabilities", in Proceedings of AFIPS, Atlantic City,NJ, AFIPS Press, pp. 483-485.

Cesta, A., Takara, A. and Moscheto, D. (2011) "Leveraging diverse regression approaches and heterogeneous machine data in the modeling of computer systems performance," in Proceedings of MSV, Las Vegas, Nevada, USA, pp. 201–207.

Cesta, A., Silva, G. and Storch, M. (2012) "Performance Prediction for Processors and External Resources," in Proceedings of CMG, Las Vegas, Nevada, USA.

Gunther, N. (2010) "Guerrilla Capacity Planning: A Tactical Approach to Planning for Highly Scalable Applications and Services". Springer.

Gustafson, J. L. (1988) "Reevaluating Amdahl's Law", in Communications of the ACM, vol. 31, no. 5, pp. 532-533.

Sun, X. and Ni, L. (1993) "Scalable problems and memory-bounded speedup", in Journal of Parallel and Distributed Computing.