Evaluation of a Dynamic Processor Overclocking Implementation for Power-Constrained Systems

Power and thermal constraints are proving to be limiting factors for computer system performance. Current approaches set the nominal frequency of a system to meet the timing, power, and cooling limits of the system, while running a worst-case workload. This limits the system to operate at this frequency even for workloads that have lower power and cooling requirements. As a consequence real workloads pay a performance cost for operating at a design point defined by the worst-case workloads.

We propose Dynamic Processor Overclocking (DPO) as a way of reducing the performance impact of the chosen power and thermal design point on workloads that have suficient slack in their power and cooling requirements. Dynamic Voltage and Frequency Scaling (DVFS) has been widely studied in recent years, while solutions exploiting it have focused on saving energy, DPO uses this technique to obtain increased performance. To do so, the DPO module monitors the workload behavior and adjusts the frequency of the system at run-time to improve performance while maintaining the power below the designed limit.

The paper presents a prototype DPO implementation and evaluates its benefits using the SPEC CPU2000 benchmarks. We identify the workload characteristics that affect the speedups obtained from overclocking as well as how the power constraints affect the performance improvement we obtain from DPO.