Dynamic voltage and frequency scaling based on workload decomposition

Abstract

This paper presents a technique called "workload decomposition" in which the CPU workload is decomposed in two parts: on-chip and off-chip. The on-chip workload signifies the CPU clock cycles that are required to execute instructions in the CPU whereas the off-chip workload captures the number of external memory access clock cycles that are required to perform external memory transactions. When combined with a dynamic voltage and frequency scaling (DVFS) technique to minimize the energy consumption, this workload decomposition method results in higher energy savings. The workload decomposition itself is performed at run time based on statistics reported by a performance monitoring unit (PMU) without a need for application profiling or compiler support. We have implemented the proposed DVFS with workload decomposition technique on the BitsyX platform, an Intel PXA255-based platform manufactured by ADS Inc., and performed detailed energy measurements. These measurements show that, for a number of widely used software applications, a CPU energy saving of 80% can be achieved for memory-bound programs while satisfying the user-specified timing constraints.

Package-Aware Scheduling of embedded workloads for temperature and Energy management on heterogeneous MPSoCs

“…We compute temperature dependence using the model introduced in [20] with the same constants mentioned in the paper for 65nm. The overhead of switching to a new frequency is 500 µs [21], [22].…”

Shadow Replication: An Energy-Aware, Fault-Tolerant Computational Model for Green Cloud Computing

“…We assume that the job is evenly partitioned among tasks and each task processes the same workload, W , measured in the number of CPU cycles [13,14]. Consequently, baring failures, tasks are expected to complete at the same time.…”

Iso-Energy-Efficiency: An Approach to Power-Constrained Parallel Computation

“…At the system level, the theoretical sequential execution time for an on-chip/off-chip workload comp rises three components [35,36]: co mputation time (with on-chip instruction execution frequency), main memo ry access latency , and I/O access time (with off-chip instruction execution frequency). Thus the theoretical execution time can be exp ressed as :…”