Interview October 2012

Oct 01 2012

Award-winning Entropy system

Interview with Jean-Marc Menaud is the project Leader of work package 4 Run, OpenCloudware.

How would you present the OpenCloudware project?

The main focus of the OpenCloudware project is to develop an open source PaaS (Platform as a Service) solution for private, public and hybrid clouds running across different Infrastructure as a Service (IaaS) models. OpenCloudware is leveraging its own IaaS model that can be seen as a complementary solution to well known open source IaaS environments such as OpenStack and OpenNebula. 

Can you explain the different cloud deployment models?

A private cloud is typically an internal cloud within an enterprise, while a public cloud is an external cloud that could be used by many companies at the same time. A hybrid cloud could go either way. If you are a developer, a hybrid cloud enables you to take an internal cloud solution to an enterprise or take an external cloud solution to many enterprises like Amazon CloudFormation does. 

What is your role in the OpenCloudware project?

As the lead of the work package 4 called Run, I am focused on developing the system that provides custom virtual machines (VMs) for the OpenCloudware project. More precisely, I am currently working on a system that is responsible for executing VMs on a cluster with SLA constraints. In this cluster, we need to find a correct position for those VMs provided by other work packages. I have been working on what used to be known as the Entropy system but which recently became known as btrPlace. We use this component in btrCloud, a software solution to manage VMs in cluster environments. More precisely, I am now working on an optimization program to reduce energy consumption in large-scale distributed systems.

The Entropy system won a prestigious prize in 2009. Can you tell us about the system?

Entropy is an open-source autonomous VM manager for datacenters. It continuously adapts the placement and the state of VMs depending on high-level constraints expressed by system administrators and applications administrators. Entropy is dynamic as it improves continuously the placement of the VMs. For the development of the Entropy system, the French Business Confederation awarded a sustainable-development prize to me and my collaborator Associate Professor Fabien Hermenier (Inria) in 2009. Entropy defines a VM in the cluster to realize optimization (see diagram below). For optimization, we can consume less power by defining the minimal amount of servers that can be running on all VMs. After that, we can shut down those servers that are not being used to save energy.

What key innovation do you bring or help to develop?

For my point of view, we have two challenges. First, we have a technical issue, how to make use of different solutions such as OpenStack, OpenNebula, and VMware. The goal is to be able to use one of these third-party solutions in a cloud that can then be combined with a private cloud. We want to define and run VMs inside clusters with  different IaaS. Second, we have a scientific challenge. When you have a distributed application you want some SLA guarantees. Typically the system must be fault tolerant. The SLA can be complied with by having two VMs for the same service. But these two VMs cannot run on the same server. So in the system you need to propose a constraint: you place a VM as you would anywhere in the cluster, but the second VM cannot be running on the same server as the first. This feature cannot be found in OpenStack or OpenNebula, so it is an innovative solution that we are addressing.

A word about yourself and your organization?

Jean-Marc-Menaud.jpgI am a researcher at Armines, a research center attached to the Ecoles des Mines in Nantes, with a staff of more than 550 employees. My main interests are in large-scale autonomic computing and systems that have high complexity both in terms of size and in terms of heterogeneity of resources and software. I'd like to focus specifically in the investigation of the required components, including language and runtime, for the implementation of complex and large-scale autonomic systems. For example, I work with the DSL programming approach for the implementation of self-optimization of autonomic system in a virtualized environment.

About Jean-Marc Menaud

An assistant professor at the Ecole des Mines de Nantes, Jean-Marc Menaud defended his Ph.D. thesis in Computer Science at the University of Rennes 1 in January 2000. Jean-Marc worked within a research group at IRISA/INRIA on cache cooperative system for large scale distributed information systems. Jean-Marc is focused on the dynamic evolution of applications, having developed the first runtime aspect weaver called Arachne for the C language. His main contributions are in the operating system domain in order to improve performance and security and also in the language domain to define new abstractions and improve the expressiveness of aspect languages.

This wiki is licensed under a Creative Commons 2.0 license
XWiki Enterprise 5.4.6 - Documentation

Site maintained by