Processor performance modeling using regression method
Joseph A. Issa · 2016
Estimating and modeling processor performance for a given workload is important for many processor architects and benchmark developers. Developing a performance estimation method using analytical approach to estimate processor performance reduces the burden of relying on simulation approach which requires a simulator and trace sampling for a given benchmark. In this paper, we present a processor performance model using regression approach rather than simulated approach. The baseline for the model is implemented using SPEC CPU2006 benchmark with Nehalem processor. The model enables the changing of different processor micro-architecture parameters such as number of cores, processor core frequency and memory DIMM speed to estimate the CPU2006 performance for a different processor of similar architecture. The model is verified to estimate performance with 10% error margin between estimated and measured baseline.