Server simulators, also known as dummy loads, are used to simulate real IT equipment for performance examining of data centers. Without these tests, potential failures could result in significant losses. Due to oversimplified designs and lack of evaluation methods, commercial dummy loads often fail to reflect accurately the true conditions of the real servers in data centers. In this study, a novel dummy-chip server simulator (DCSS) is developed, which can well reproduce the working mechanism of real servers. In this paper, three indices are proposed for the first time to evaluate the performance from different perspectives, i.e. Coefficient of Heat Removal (CHR) for heat removal capability, Coefficient of Temperature Deviation (α) for temperature rating, and Power Usage per Computing (PUC) for energy and computation efficiency. Experimental tests are conducted in this study, and results show that the good performance of the DCSS approaching the real server, and the values of CHR, α, and PUC indicate significant improvements compared to the commercial dummy load. The findings of this study present straightforward guidance for IT server simulator design, and demonstrate the advancement of DCSS and proposed index for data center performance evaluation.
Read full abstract