Since September 2005, I have been a PostDoctorant at the Oak Ridge National Laboratory (ORNL), USA. I carry on cluster and grid computing for High Performance Computing (HPC), focusing of virtualization, Operating System (OS), and High Availability (HA) issues.
My position at the Oak Ridge National Laboratory allows me to work on two FastOS projects (funded by the Department Of Energie): the PetascaleSSI project and the MOLAR project. I am also Investigator in the LDRD project (funded by the Oak Ridge National Laboratory) Virtualized System Environments for Petascale Computing and Beyond.
In 2005 and 2006, I have also been technical expert at the European Commission for the review of a project funded by the European Commission.
From March 2004 to September 2005, I had been an industrial PostDoctorant at INRIA, co-funded by Électricité De France Research and Development (EDF R/D). I carry on methods for building, programming, and using clusters, developing a fully integrated and easy to install software bundle designed for high performance cluster computing. A solution to offer ease programming and ease use of clusters is to used a Single System Image (SSI) as cluster system. So, I work on the creation of a SSI package for the OSCAR, a cluster distrubution which provides a snapshot of the best known methods for building, programming, and using clusters. For that, I have integrated the OSCAR team at the Oak Ridge National Laboratory, Oak Ridge, Tennessee, USA, during one year in the framework of my PostDoctorant position. I also worked on the OSCAR on Debian project, a port of the OSCAR distribution on the Debian Linux distribution.
I have made my Ph.D. at the University of Rennes 1, funded by Électricité De France Research and Development (EDF R/D). During my Ph.D., I carry on my research activities at IRISA/INRIA in the PARIS project-team (http://www.irisa.fr/paris). I worked on global scheduling and global process management in Kerrighed, an operating system for high performance computing on clusters.
Description: This research project intends to address scalability, manageability, and ease-of-use chal- lenges in petascale system software and application runtime environments through the development of a virtual system environment (VSE). In addition to providing a scalable and reliable "sandbox" environment for scientific application development on desktops and clusters, the VSE will offer an identical production environment for scientific application deployment on terascale and petascale HEC systems. The VSE concept enables "plug-and-play" supercomputing through desktop-to-cluster-to-petaflop computer system-level virtualization based on recent advances in hypervisor virtualization technologies. The overall goal of this effort is to advance the race for scientific discovery through computation by enabling day-one operation capability of newly installed systems and by improving productivity of scientific application development and deployment.
Description: This project is a multiinstitution research effort that concentrates on adaptive, reliable, and efficient operating and runtime system solutions for ultra-scale high-end scientific computing on the next generation of supercomputers. It addresses the challenges outlined by the FAST-OS - forum to address scalable technology for runtime and operating systems - and HECRTF - high-end computing revitalization task force - activities by providing an adaptable runtime support for high-end computing operating and runtime systems. This research primarily concentrates on advancing computer reliability, availability and serviceability (RAS) management systems to run large and long-running applications efficiently on future ultra-scale computers, and on providing advanced monitoring and adaptation mechanisms for improved application performance and predictability. For more information, please visit www.fastos.org/molar.
Description: Peta-scale computers with thousands of times more computational power will be available to scientists by the end of this decade. The research and development of these next generation computing architectures and the corresponding computing environments will accelerate scientific discoveries within the Office of Science as well as within the high-end computing community in general. The overwhelming size and complexity of a peta-scale system will require a computing environment that addresses the scalability of the file system, a network that facilitates communications across the scale of processors, and an aggressive approach for dealing with operating system noise. We believe the best way to reach this environment is to scale a single system image Linux environment to 100,000 processors. We propose fundamental research and development using the Open Single System Image (OpenSSI) project as a baseline to provide a balanced solution to accomplish the goals outlined in this proposal. OpenSSI is an open source project maintained by Bruce Walker (Co-PI of this proposal) which provides a single root file system and single process space across distributed resources. OpenSSI is a full implementation of the single system image and has a more fault tolerant peer-to-peer communication system than other implementations available.
Description: Today, clusters are widely used to compute a wide range of high performance scientific applications which may be sequential or parallel. Previous studies show that different kinds of application workloads and cluster usages imply different needs in resource management, and thus different scheduling policies. In my thesis work I have proposed a modular architecture to provide an adaptive global scheduler and a development framework of new policies allowing the integration in Kerrighed of traditional scheduling policies as well as any new policy. The proposed global scheduler is based on mechanisms for dynamic configuration and hot-load / hot-eviction of scheduling modules. A development toolkit for scheduling policies has also been implemented to simplify the development of new policies. The global scheduler is based on mechanisms to efficiently manage processes in a cluster. I have proposed a mechanism of \textit{ghost process} allowing the implementation of process migration, remote process creation and process checkpoint and restart. This mechanism takes advantage of the Kerrighed's distributed shared memory for thread management. Based on these mechanisms, the Kerrighed operating system provides a pthread interface, allowing in particular the execution of OpenMP applications on a cluster. The process checkpoint/restart mechanism has been used in backward error recovery protocols for parallel applications, designed in collaboration with Ramamurthy Badinath, Associate Professor at the Indian Institute of Technology of Kharagpur. A coordinated checkpointing strategy for shared memory parallel applications has been implemented in the Kerrighed operating system. All the mechanisms and algorithms that I have proposed have been integrated in a prototype of Kerrighed operating system. Kerrighed is an open source software available at the following URL http://www.kerrighed.org. My work was validated by the experimentation of industrial applications provided by EDF R/D on top of Kerrighed. The Kerrighed project is now lead and developed by the KerLabs company (http://www.kerlabs.com/).
The list of my publications is available here: https://www.csm.ornl.gov/srt/people/gvallee/publications/index.html.