Modeling energy consumption

on June 28, 2016 at 9:30 am

List of presentations :

================ Jalil Boukhobza (UBO, Lab-STICC)

Title: “A methodology for evaluating the performance and energy consumption of embedded flash memory-based storage systems”.

Abstract: We will begin this presentation by introducing our team’s work in the field of storage and, more specifically, energy consumption. We will then describe a work carried out by Pierre Olivier (former PhD student of the team) which represents a common approach in the team and which targeted the evaluation of the performance and energy consumption of the secondary storage service in an embedded operating system using NAND flash memory. The contributions made in this work revolve around a modeling methodology for estimating the performance and power consumption of FFS (Flash File System) embedded storage systems. This methodology is divided into three phases. In the exploration phase, we use micro-benchmarks to identify the elements of the storage system that have an impact on performance and energy consumption. In the modeling phase, this impact is represented in the form of models of various types (e.g. functional, performance and consumption models). In the simulation phase, the models are implemented in a simulator, enabling estimates to be made of the performance and power consumption of a flash-based storage system subjected to a given I/O load.

================ Julien Forget (Univ. Lille, CRIStAL)

Title: “Towards an energy consumption model for embedded real-time software”.

Abstract: We will present work carried out as part of several successive IRCICA projects aimed at modeling the energy consumption of embedded software on modest-sized embedded boards. As part of this work, we have developed experimental tools and a series of small applications of various kinds. This enabled us to highlight the significant variability in energy consumption depending on the type of software being run.

================ Abdoulaye Gamatié (LIRMM)

Title: Dealing with Energy-Efficiency in Next-Generation Compute Systems

Abstract: The recently observed convergence of embedded and high-performance computing lies in the growing energy/performance requirement in both domains. This trend has a heavy impact on the way scientific problems and industry challenges must be addressed. One attractive solution consists of a design continuum that seamlessly goes from software level to hardware technology level, via hardware architecture level. Indeed, power saving opportunities exist at each of these levels, and real measurable gains will come from the synergistic focus on all these levels at once. This talk advocates a cross-disciplinary approach for devising adequate solutions to the energy-efficiency challenge in next-generation compute systems.

================ Kevin Marquet (INSA Lyon)

Title: “Evaluation of the consumption of a sensor carrying the NVRAM”

Summary: We have developed an embedded system hardware/software protype capable of running on intermittent power. The approach is based on a backup of data in NVRAM. I will present the prototype, how we simulate this platform, and the limitations of this simulation.

================ Romain Rouvoy (Univ. Lille, CRIStAL)

Title: “PowerAPI: Building Software-Defined Power Meters”

Summary: This presentation summarizes the work we have undertaken in recent years on real-time consumption measurement at different granularities. In particular, we present here the PowerAPI library, a toolkit for building software wattmeters. These software wattmeters require no third-party hardware and are able to estimate in real time the consumption of different components (CPU, disk, etc.) at different scales (machine, processes, containers, classes, methods) and for different architectures (Intel, AMD, ARM). Adopting a middleware approach allows us to integrate software wattmeters designed with PowerAPI into different contexts and share metrics through different communication channels.

================

IRCICA