Optimizing power allocation to CPU and memory subsystems in overprovisioned HPC systems

Abstract

Abstract-Energy consumption and power draw pose two major challenges to the HPC community for designing larger systems. Present day HPC systems consume as much as 10MW of electricity and this is fast becoming a bottleneck. Although energy bills will significantly increase with machine size, power consumption is a hard constraint that must be addressed. Intel's Running Average Power Limit (RAPL) toolkit is a recent feature that enables power capping of CPU and memory subsystems on modern hardware. In this paper, we use RAPL to evaluate the possibility of improving execution time efficiency of an application by capping power while adding more nodes. We profile the strong scaling of an application using different power caps for both CPU and memory subsystems. Our proposed interpolation scheme uses an application profile to optimize the number of nodes and the distribution of power between CPU and memory subsystems to minimize execution time under a strict power budget. We validate these estimates by running experiments on a 20-node (120 cores) Sandy Bridge cluster. Our experimental results closely match the model estimates and show speedups greater than 1.47X for all applications compared to not capping CPU and memory power. We demonstrate that the quality of solution that our interpolation scheme provides matches very closely to results obtained via exhaustive profiling.

Analyzing the Impact of CPU Pinning and Partial CPU Loads on Performance and Energy Efficiency

“…• In other research, the shared resource is the power budget-Raghavendra et al [11] explain why coordinated power management is necessary and investigate a wide range of coordinated power management scenarios, Sarood et al [25] show how using more computers with limited power can benefit energy efficiency of certain workloads, Subramaniam et al [1] use power limits to improve energy proportionality. Unfortunately, the very high and still growing number of possible resource interactions does not lend itself to comprehensive characterization through individual resource models.…”

Maximizing Throughput of Overprovisioned HPC Data Centers Under a Strict Power Budget

“…Earlier work [3], [4] shows that an increase in the power allowed to the processor (and/or memory) does not yield a proportional increase in the application's performance. As a result, for a given power budget, it can be better to run an application on larger number of nodes with each node capped at lower power than fewer nodes each running at its TDP.…”

“…As a result, for a given power budget, it can be better to run an application on larger number of nodes with each node capped at lower power than fewer nodes each running at its TDP. The optimal resource configuration for an application can be determined by profiling an application's performance for varying number of nodes, CPU power and memory power and then selecting the best performing configuration for the given power budget [4]. In this work, we address the data center scenario in which an additional decision has to be made: how to distribute available nodes and power amongst the queued jobs.…”

Energy-efficient computing for HPC workloads on heterogeneous manycore chips

“…This is also called overprovisioning. In our previous work ( [41,42]), we have shown significant improvement in performance of a data center by using overprovisioning under a strict power budget. We have also shown the benefit of using integer linear programming methods for improving the performance of applications on chips with low voltage operation under a strict power budget [47].…”