A simple GRID system model for estimating a job scheduling algorithm impact on power consumption

UDC 004.051

I.L. Kaftannikov, South Ural State University, Chelyabinsk, Russian Federation, kil@comp.susu.ac.ru

K.M. Khankin, South Ural State University, Chelyabinsk, Russian Federation, hc@comp.susu.ac.ru

Abstract

The model of GRID system including two clusters is described. Model is created by simulation in AnyLogic. The main purpose of the model is to estimate the impact of GRID broker scheduling algorithm on GRID power consumption. Model displays task scheduling process and counts average power consumption by task, average local queues length, average waiting time in queues and probability distribution for average waiting time. The scheduling algorithm that makes power consumption lower is described and compared with algorithm of queue length equality. Described algorithm performs scheduling by the criterion of graphic processing unit presence on the cluster and implementation for the GPU presence in task.

Full text

Keywords

grid, power consumption, scheduling algorithm

References

1. Khankin K.M. Efficiency Comparsing of OpenMP, nVidia CUDA and StarPU in the Example of Matrix Multiplication [Sravnenie effektivnosti tekhnologiy OpenMP, nVidia CUDA i StarPU na primere zadachi umnozheniya matrits]. Vestnik YUURGU, 2013, Vol. 13, № 1, pp. 34–41.
2. Review: Intel Xeon E5-2600 Eight-Core CPUs / Tom's IT Pro. – http://www.tomsitpro.com/articles/xeon-e5-2687w-benchmark-review-cores,2-288-11.html
3. TESLA™ C2050 / C2070 GPU Computing Processor. Supercomputing at 1/10th the cost. / NVIDIA Corporation. – www.nvidia.com/docs/IO/43395/ NV_DS_Tesla_C2050_C2070_ jul10_lores.pdf

Source

Bulletin of the South Ural State University. Ser. Computer Technologies, Automatic Control, Radio Electronics, 2013, vol. 13, no. 2, pp. 131-137. (in Russ.) (Brief Reports)