Hardware support for accurate per-task energy metering in multicore systems

Abstract

Accurately determining the energy consumed by each task in a system will become of prominent importance in future multicore-based systems because it offers several benefits, including (i) better application energy/performance optimizations, (ii) improved energy-aware task scheduling, and (iii) energy-aware billing in data centers. Unfortunately, existing methods for energy metering in multicores fail to provide accurate energy estimates for each task when several tasks run simultaneously. This article makes a case for accurate Per-Task Energy Metering (PTEM) based on tracking the resource utilization and occupancy of each task. Different hardware implementations with different trade-offs between energy prediction accuracy and hardware-implementation complexity are proposed. Our evaluation shows that the energy consumed in a multicore by each task can be accurately measured. For a 32-core, 2-way, simultaneous multithreaded core setup, PTEM reduces the average accuracy error from more than 12% when our hardware support is not used to less than 4% when it is used. The maximum observed error for any task in the workload we used reduces from 58% down to 9% when our hardware support is used. ACM Reference Format:Liu, Q., Moreto, M., Jimenez, V., Abella, J., Cazorla, F. J., and Valero M. 2013. Hardware support for accurate per-task energy metering in multicore systems. ACM Trans.

Power monitoring with PAPI for extreme scale architectures and dataflow-based programming models

“…In such a scenario, each measurement will correspond to power consumed by different segments of tasks executing on the different cores of a socket. Thus, when all measurements have been collected, inferring the average power consumption of each thread type requires solving a system of linear equations [29].…”

Time and energy modeling of a high-performance multi-threaded Cholesky factorization

“…Few past efforts have considered modeling the role of energy [4]. In most cases, these models characterize energy consumption by breaking the application into microinstructions [5][6][7][8]. Here, we follow a higher level approach which considers coarse-grain operations and does not need to rely on architecturespecific hardware counters.…”

Sensible Energy Accounting with Abstract Metering for Multicore Systems

“…Power models rely on collecting data from a set of PMCs, and voltage and temperature information, to estimate power through correlation. -Per-Task Energy Metering (PTEM) [Liu et al 2013[Liu et al , 2014] estimates the energy actually consumed by each application simultaneously running in a multicore system. The main challenge of PTEM is dealing with shared hardware resources, as the energy consumption of applications significantly changes depending on the co-running applications.…”

“…Let us illustrate the concept of SEA and how it differs from PTEM with an example. We simulate several SPEC CPU 2006 benchmarks on a 4-core multicore architecture comprising a shared last-level cache (LLC) 2 and the PTEM technique [Liu et al 2013]. We choose namd, astar, and libquantum benchmarks since they have different (LLC) utilization levels.…”