An Improved Stochastic-based Approach for Optimizing Energy-Time Tradeoff in Multicore Systems

Nguyễn, Quang Vinh; Hajiamini, Shervin; Shirazi, Behrooz

doi:10.1109/igsc51522.2020.9290859

Cited by 1 publication

(1 citation statement)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Then we proceed with a numerical multi-objective analysis by proposing two fast and approximate algorithms for the performance and power consumption trade-off. To model DVFS system, most analytical methods are based on stochastic modeling as: in (He et al 2020) a Petri Net model is proposed for dynamic scaling and VM migration in Energy-Aware cloud system; in (Dargie 2015) a random variable analysis is set for estimating the relationship between workload and power consumption in multi-core processor; while in (Nguyen et al 2020) a Hidden Markov Model is used with a predicting algorithm for hidden states of the system applied to a multi-core DVFS for energy-time trade-off; Markovian Decision Process (MDP) also is proposed (Anselmi et al 2021), for models with tasks deadlines and unbounded state space and speed rates. These models often suffer from the explosion of the set of states.…”

Section: Introductionmentioning

confidence: 99%

Stochastic Modeling And Optimization For Power And Performance Control In DVFS Systems

Ait El Mahjoub,

Le Corre,

Castel-Taleb

2023

ECMS 2023 Proceedings Edited by Enrico Vicario, Romeo Bandinelli, Virginia Fani, Michele Mastroianni

View full text Add to dashboard Cite

The paper addresses the problem of performance-energy trade-off in DVFS (Dynamic Voltage Frequency Scaling) systems. We propose a stochastic hybrid model between hysteresis models and server block models. We provide a closed form for the steady-state distribution probability and we establish a "st" type order to compare the performance measures. The fast computation of power and performance measures leads to a multi-objective optimization analysis in two forms: a scalarization method and a Pareto based method. For the two approaches, we propose fast and efficient approximate algorithms that construct progressively an optimal solution. To discuss results, the model is used to simulate a physical server hosting several VMs (Virtual Machines) where we investigate optimal thresholds for the performance-energy trade-off.

show abstract