CloudPack

Abstract

HAL is a multidisciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L'archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d'enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

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

“…Importantly, the results from the worst-case scenarios cannot be simply "scaled down" to reflect partial loads, because they lack information about power usage and the corresponding application performance. While power usage may be modeled as a function of utilization [6,[9][10][11] or frequency [12], the relation between CPU utilization and application performance is not necessarily linear and potentially complex and difficult to predict [13]. For example, Martinec et al [14] have shown that the throughput of the ActiveMQ JMS broker can more than double when CPU utilization increases from 80% to 95%, with the choke point still safely away at almost three times the throughput at 80% utilization.…”

“…When both criteria fail, the corresponding number is typeset in italics. Failures of only the first (symmetry) or the second (p-value) criterion are represented by sans-serif and standard font, respectively 9 . Results for which both criteria are met are typeset in bold.…”

“…Chase et al [18] provide one of the early examples, pitting the revenue expressed in terms of customer utility functions against the cost of resources. Among more recent research results, Ishakian et al [9] show that expressing application resource requirements in terms of flexible preferences rather than fixed demands can reduce consolidated resource waste by up to 60%. Another economically motivated research direction is represented by Wang et al [19], who develop a model for properly attributing costs in consolidated workload scenarios.…”

“…Coarse granularity simulation often uses processor utilization log replay to approximate realistic workloads, assuming additive character of processor utilization. When considered, power usage is modeled for example as a function of utilization [6,[9][10][11] or frequency [12]. Here, our work can contribute to a more realistic evaluation with an empirical model for deriving processor utilization and power consumption under workload colocation conditions-as we show, true utilization and consumption values on contemporary platforms can substantially differ from the results of existing models.…”