Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Green Adaptation of Real-Time Web Services for Industrial CPS within a Cloud Environment

Teresa Higuera, José L. Risco-Martín, Patricia Arroba, José L. Ayala

TL;DR

The paper addresses energy-efficient, real-time provisioning for industrial CPS running in cloud environments under hard timing constraints. It introduces a leakage-aware power model that ties static power to temperature and DVFS-driven dynamic power, enabling accurate energy estimation in data centers. A multi-objective optimization using NSGA-II jointly tunes DVFS settings and workload allocation to minimize energy while controlling deadline misses, with a reservation-based scheduling perspective. Experimental results on real hardware (Intel Xeon E5620 and AMD Opteron 270) demonstrate feasible, energy-diverse Pareto fronts and superior energy-efficiency compared to baselines, highlighting practical relevance for green CPS in the cloud. The work paves the way for deeper integration of criticality-aware scheduling and virtualization in energy-conscious real-time cloud services.

Abstract

Managing energy efficiency under timing constraints is an interesting and big challenge. This work proposes an accurate power model in data centers for time-constrained servers in Cloud computing. This model, as opposed to previous approaches, does not only consider the workload assigned to the processing element, but also incorporates the need of considering the static power consumption and, even more interestingly, its dependency with temperature. The proposed model has been used in a multi-objective optimization environment in which the Dynamic Voltage and Frequency Scaling (DVFS) and workload assignment have been efficiently optimized.

Green Adaptation of Real-Time Web Services for Industrial CPS within a Cloud Environment

TL;DR

The paper addresses energy-efficient, real-time provisioning for industrial CPS running in cloud environments under hard timing constraints. It introduces a leakage-aware power model that ties static power to temperature and DVFS-driven dynamic power, enabling accurate energy estimation in data centers. A multi-objective optimization using NSGA-II jointly tunes DVFS settings and workload allocation to minimize energy while controlling deadline misses, with a reservation-based scheduling perspective. Experimental results on real hardware (Intel Xeon E5620 and AMD Opteron 270) demonstrate feasible, energy-diverse Pareto fronts and superior energy-efficiency compared to baselines, highlighting practical relevance for green CPS in the cloud. The work paves the way for deeper integration of criticality-aware scheduling and virtualization in energy-conscious real-time cloud services.

Abstract

Managing energy efficiency under timing constraints is an interesting and big challenge. This work proposes an accurate power model in data centers for time-constrained servers in Cloud computing. This model, as opposed to previous approaches, does not only consider the workload assigned to the processing element, but also incorporates the need of considering the static power consumption and, even more interestingly, its dependency with temperature. The proposed model has been used in a multi-objective optimization environment in which the Dynamic Voltage and Frequency Scaling (DVFS) and workload assignment have been efficiently optimized.
Paper Structure (17 sections, 10 equations, 4 figures, 4 tables)

This paper contains 17 sections, 10 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. The industrial services execution model
  4. Task characterization
  5. The task model
  6. Hard real-time tasks
  7. Soft real-time tasks
  8. Feedback control tasks
  9. The parallel scheduling
  10. Power and energy model
  11. Leakage power
  12. Dealing with DVFS
  13. Energy model
  14. Multi-Objective Optimization Algorithm
  15. The solver
  16. ...and 2 more sections

Figures (4)

  • Figure 1: An overrun in response time (i.e., a deadline miss) has a different value function depending on its possible consequences
  • Figure 2: Chromosome encoding
  • Figure 3: Pareto front obtained with NSGA-II after optimizing the allocation of tasks over the Intel architecture.
  • Figure 4: Pareto front obtained with NSGA-II after optimizing the allocation of tasks over the AMD architecture.