3. Impact (1 page maximum)
Indiquez l'impact de cette collaboration sur :
- les objectifs scientifiques des équipes participantes ;

One of the goal primary goal of the collaboration is to generate scientific results in a larger context: instead of focusing on a specific platform, we try to analyze challenges related to HPC in a broader context, targeting all major HPC platforms. We also plan to conduce those studies with real applications (including applications from ORNL) instead of micro-benchmarks. That will allow us to focus on the user point of view, and be sure we address user issues.

- les relations entre les partenaires et entre les instituts (par exemple discutez de la complémentarité, de la similarité pour un effet de masse critique, de la répartition des tâches pour un gros développement, etc.)

The SRT team focuses on clusters and parallel large-scale system such as Cray systems; while the PARIS project-team is expert in grid computing. The collaboration between the two teams is therefore complementary and will allow the “équipe associée” to impact all HPC platforms and federate the community efforts.

It also means that ORNL will focus on studies targeting clusters and large-scale parallel systems while the PARIS project-team will focus on clusters and grids. The different efforts will be organized accordingly.

3.1. sur la collaboration deja existante avec votre partenaire

The creation of a "équipe associée" between the SRT team at ORNL and the PARIS team at INRIA will foster the study and implementation of system-level virtualization solutions for HPC. The goal of such a collaboration is to study the usage and implementation of operating systems and their associated tools in distributed or parallel platforms in the context of HPC. Doing so, the two teams will propose new standards to the international community both in the domain of grid and cloud computing, and of large-scale parallel systems (such as IBM and Cray machines).

This effort will initially be based on the ongoing studies on system-level virtualization for HPC. Those studies focused on the formal definition of virtualization and on the study of the usage of virtualization in conjunction with SSI techniques. Those studies can therefore be followed by some deeper studies in different domains. For instance: How can we use system-level virtualization in the context of cloud computing? What are the needed virtualization services? How can we implement those services? What are the common points between the different HPC platforms in term of virtualization capabilities?

We ultimately target the following capabilities:

• Transparency: users should be able to use distributed systems the same way they use a standard system. For that, new tools must be implemented and current tools extended, extending the capabilities of normal Unix-/Posix-like tools. This is mandatory for hiding the increasing complexity of HPC systems.

• Adaptability: modern HPC platforms face two critical challenges, (i) how can we adapt a given system to a specific platform (platforms differ deeply in term of characteristics on the hardware aspects)?, and (ii) on a given platform, how can we adapt the system to configuration modifications at runtime (failures for instance).

• Resiliency: because of their scale, modern platforms have to deal with many failures. To guarantee non-stop computing at the application level, the system must include fault tolerance or fault avoidance mechanisms and policies. This includes adaptability capabilities but also fault tolerance or avoidance mechanisms (which can be based on virtualization, e.g., migration of virtual machines).

The collaboration will result in common open source software (available for the international community) and also in scientific publications submitted to the major international conferences (such as IEEE/ACM SuperComputing, ISC, ACM EuroSys, IEEE/ACM EuroPar, IEEE/ACM IPDPS, ACM VEE, IEEE CCGRID and IEEE Cluster).

3.2 sur la collaboration avec d'autres projets INRIA

Browsing the INRIA website, we did not find any previous collaboration between ORNL and INRIA, except for collaborations between various INRIA project-teams (e.g., GRAAL, RESO, GRAND-LARGE) and Jack Dongarra who is co-affiliated to ORNL.

Several INRIA project-teams are working on system-level virtualization problematics and may be interested in collaborating with the SRT and PARIS team: the ASCOLA project-team (with the Entropy prototype, a consolidation manager for clusters), the RESO project-team (network virtualization), the MESCAL project-team (with the SAMORY prototype which is an architecture to provide resiliency to parallel applications running on top of virtual clusters).

3.3 sur la collaboration avec d'autres equipes de l'organisme etranger

It is also expected that this collaboration will foster new collaborations with other ORNL groups in the area of operating systems for HPC and system tools.

Specifically, the System Research Team works closely with the Tool group at ORNL, led by Richard L. Graham. This group aims to define and develop the next generation tools, targeting peta-scale platforms and behind. The two major constraints for such tools are scalability and resiliency.

The SRT team also collaborates with several universities in the USA, creating collaboration opportunities for the PARIS project team. For instance, the SRT team currently have active collaborations with University of New Mexico (on the topic operating systems for HPC - virtualization), Northwestern University (on the topic operating systems for HPC - virtualization), LATech (on the topic of resiliency), and North Carolina University (on the topic of operating systems for HPC and resiliency).

4. Divers : toute autre information que vous jugerez utile d'ajouter.

II. PREVISIONS 2009

Programme de travail

In 2009, we plan to focus on three different topics and the organization of an event specific to the collaboration. We plan to work on system tools, operating systems, and resiliency challenges. Those challenges are actually not disconnected, they should converge in a single software solution for HPC in the future.

* System Tools:

Identification of common capabilities between system tools developed at INRIA (the Aladdin software stack and some research prototype based on this software stack for the management of virtual machines) and those developed at ORNL (OSCAR and more precisely the OSCAR-V package [8]). The goal of this effort is to try to identify common mechanisms and to try to define "standards" for such tools, discussing with the international community on the subject. This will focus on the "user point-of-view": how can we simplify the user (application user) life when using distributed or parallel platforms, i.e., the descriptions of application's needs in term of software stack and execution environment, and the description of needed resource. Underneath, the system tools will find and allocate requested resources, prepare the execution environment dynamically, deploy it and execute the application. The tool will have to be functional on the three target platforms: clusters, Cray-like systems and Grids.

This effort will lead to the release of a version of OSCAR/OSCAR-V that can generate Aladdin images for a deployment on top of Aladdin. This release will allow users to describe the execution environment described by the users and automatically deploy the environment on the target platform. In other terms, the goal is to implement the notion of virtual platform and evaluate its usage on the three target platforms.

* Operating Systems:

The execution of application on Grids is based on the concept of Virtual Organization (VO) which is the implementation of user access policies, resources allocation of shared resources, and deployment and management of applications [19,22,35]. At the other end, physical resources must be assigned to users, via VOs. This could be done using virtual machines that isolate the user from the bare hardware (security) and provide a execution environment that fits application's needs. In between, in order to extend the local view of virtual machines (virtual machines are about local resources, there is no notion of distributed platforms), we propose the definition and implementation of "virtual platforms" (VP). The virtual platform is a subset (in other terms partition) of the distributed physical resources; for instance, it can be a et of virtual machines. Therefore a VP is only the user view of the distributed platform, i.e., compared to VOs, it does not include policies about user access authorization and policies about the resources usage (accounting and so on) which are specific to the VO. Our first task for this effort is therefore to specify the software stack VO/VP/VM based on the three target platforms, with the goal of writing and submitting a white paper describing the results of the study.

* Resiliency:

The three target platforms, because of their scale, have in common failures that can impact application execution. To address this issue, resiliency policies have been implemented for all of these platforms. We propose to identify typical system policies for those platforms and analyze common points and differences. This effort aims to answer the following question: does the differences of the three target platforms impacts the nature of resiliency policies? or is it only a difference of parameters (the policy semantic remains the same; policies do not differ in nature)?

We plan to write and submit a whitepaper on the subject to an international scientific event.

* Miscellaneous:

The two teams already have several open source software prototypes that could be used as basis for the developments performed by the “équipe associcée”:

• XtreemOS: http://www.xtreemos.eu/

• Kerrighed: http://www.kerrighed.org/

• Aladdin: https://www.grid5000.fr/mediawiki/index.php/Grid5000:Home

• OSCAR: http://oscar.opensourcecluster.org/

• OSCAR-V: http://www.csm.ornl.gov/srt/oscarv/

The collaboration will lead to the implementation of new open source software or the extension of existing software.

Programme d'échanges avec budget prévisionnel

1. Echanges

Décrivez les échanges prévus dans les deux sens : invitations de chercheurs de votre partenaire et missions INRIA vers votre partenaire ;
Précisez s'il s'agit de chercheurs confirmés ou de juniors (stagiaires, doctorants, post-doctorants) ;
Motivez, si possible, les raisons scientifiques (travail commun, workshop,..) et précisez la durée prévue ;

The collaboration will be implemented via the exchange of junior and staff researchers. A Ph.D. Student, Jérôme Gallard, will visit ORNL during 3 months during spring 2009 for the design and implementation of the concept of virtual platforms. Christine Morin, the PhD advisor, will also visit ORNL for a one week period during this internship.

We plan to organize a workshop open to INRIA and ORNL partners, focusing on some of the challenges addressed by the associated team. We actually plan to organize such a workshop every year, going forth-and-back between France and the USA. The first workshop will be organized in France.

An application for a one month sabbatical period has been submitted to the University of Rennes 1 by Stephen L. Scott with the goal of spending one month during the summer 2009 (June-August period). If the application is accepted, our workshop will be organized during this period, and the fundings allocated in the framework of the SER-OS associated team to the visit of Stephen L. Scott will be reassigned for the visit of other ORNL team members or for the internship of other junior researchers from the PARIS project-team at ORNL.

Geoffroy Vallée, research scientist at ORNL, and Thomas Naughton, both Ph.D. Student at the university of Reading in the UK and research staff at ORNL, will visit the PARIS project-team for a one-week period (most certainly in December).

We also plan to meet during the SuperComputing conference organized each year in November in the United States in order to manage the project and synchronize the different efforts.

Résumez ensuite ces informations dans les tableaux 1 et 2 ci-dessous en faisant une estimation budgétaire :

2. Cofinancement

Cette coopération bénéficie-t-elle déjà d'un soutien financier de la part de l'INRIA, de l'organisme étranger partenaire ou d'un organisme tiers (projet européen, NSF, ...) ?

The XtreemOS project will provide 5.000 euros to cover travel expenses for researchers involved in XtreemOS.

The SRT team can cover a couple of travels for different meetings.

Indiquez ces éléments et donnez les montants associés. Dans le cas où votre proposition serait retenue, vous parait-il probable d'obtenir de l'organisme étranger partenaire un soutien financier symétrique ? De quel montant ?

No, but ORNL and more specifically Department of Energy has programs for summer student internships. The ORNL team has a project for 3 years, starting in 2008, titled “Reliability, Availability, and Serviceability (RAS) for Petascale High-End Computing and Beyond”, and funded by Office of Advanced Scientific Computing Research, Office of Science, U.S. Department of Energy. Program: Operating/Runtime Systems for Extreme Scale Scientific Computation (LAB 07-23).

3. Demande budgétaire

Indiquez, dans le tableau ci-dessous, le coût global estimé de la proposition et le budget demandé à la DRI dans le cadre de cette Equipe Associée (maximum 20 K€).

References

[1] Stephen L. Scott, Geoffroy Vallée, Thomas Naughton, Anand Tikotekar, Christian Engelmann, and Hong Ong. Research on system-level virtualization at the oak ridge national laboratory. Future Generation Computer Systems, 2008. To appear.

[2] Christine Morin, Pascal Gallard, Renaud Lottiaux, and Geoffroy Vallée. Towards an efficient single system image cluster operating system. Future Generation Computer Systems, 20(2), January 2004.

[3] Christian Engelmann,  Geoffroy Vallée,  Thomas Naughton, and Stephen L. Scott. Proactive Fault Tolerance Using Preemptive Migration: Model and Classification. In Proceedings of the 17th Euromicro International Conference on Parallel, Distributed and network-based Processing (PDP 2009). To appear.

[4] Kulathep Charoenpornwattana, Chokchai Leangsuksun, Geoffroy Vallée, Anand Tikotekar, and Stephen Scott. A scalable unified fault tolerance for HPC environments. In Proceegins of the 9^th LCI International Conference on High-Performance Clustered Computing, April 2008.

[5] Geoffroy Vallée, Kulathep Charoenpornwattana, Christian Engelmann, Anand Tikotekar, Chokchai Leangsuksun, Thomas Naughton, and Stephen L. Scott. A framework for proactive fault tolerance. In Proceedings of the Third International Conference on Availability, Reliability and Security (ARES 2008 - The International Dependability Conference), pages 659–664, Barcelona, Spain, March 4-7, 2008. IEEE Computer Society.

[6] Geoffroy Vallée, Stephen L. Scott, and al. System-level virtualization for high performance computing. In Proceedings of the 16th Euromicro International Conference on Parallel, Distributed and network-based Processing (PDP 2008), pages 636–643, Toulouse, France, February 13-15, 2008. IEEE Computer Society.

[7] Anand Tikotekar, Geoffroy Vallée, Thomas Naughton, Stephen L. Scott, and Chokchai Leangsuksun. Evaluation of fault-tolerant policies using simulation. In Proceedings of the 9th IEEE International Conference on Cluster Computing (Cluster), Austin, Texas, USA, September 17-20, 2007.

[8] Geoffroy Vallée, Thomas Naughton, and Stephen L. Scott. System management software for virtual environments. In CF ’07: Proceedings of the 4^th international conference on Computing frontiers, pages 153–160, New York, NY, USA, May 7-9, 2007. ACM.

[9] Jérôme Gallard, Geoffroy Vallée, Adrien Lèbre, Christine Morin, Pascal Gallard, and Stephen L. Scott. Complementarity between virtualization and single system image technologies. In 3^rd Workshop on Virtualization in High-Performance Cluster and Grid Computing (VHPC ’08), Las Palmas de Gran Canaria, Canary Island, Spain, August 2008.

[10] Anand Tikotekar, Geoffroy Vallée, Thomas Naughton, Hong Ong, Christian Engelmann, and Stephen L. Scott. An analysis of hpc benchmark applications in virtual machine environments. In 3^rd Workshop on Virtualization in High-Performance Cluster and Grid Computing (VHPC ’08), Las Palmas de Gran Canaria, Canary Island, Spain, August 2008.

[11] Geoffroy Vallée, Anand Tikotekar, Chokchai Leangsuksun, and Stephen L. Scott. Impact of fault- tolerance policies: Feasibility study. In HAPCW’08: High Availability and Performance Computing Workshop, Denver, Colorado, USA, April 3–4, 2008. Held in conjunction with High-Performance Computer Science Week (HPCSW) 2008.

[12] Anand Tikotekar, Geoffroy Vallée, Thomas Naughton, Hong Ong, Christian Engelmann, and Stephen L. Scott. Effects of virtualization on a scientific application – running a hyperspectral radiative transfer code on virtual machines. In Proceedings of the 2nd Workshop on System-level Virtualization for High Performance Computing (HPCVirt) 2008, in conjunction with the 3^rd ACM SIGOPS European Conference on Computer Systems (EuroSys) 2008, Glasgow, UK, March 31, 2008.

[13] T. Naughton, G. Vallée, and S. L. Scott. Dynamic adaptation using Xen. In Proceedings of the 1^st Workshop on System-level Virtualization for High Performance Computing (HPCVirt) 2007, in conjunction with the 2^nd ACM SIGOPS European Conference on Computer Systems (EuroSys) 2007, Lisbon, Portugal, March 20, 2007.

[14] C. Engelmann, S. L. Scott, H. Ong, G. Vallée, and T. Naughton. Configurable virtualized system environments for high performance computing. In Proceedings of the 1st Workshop on System-level Virtualization for High Performance Computing (HPCVirt) 2007, in conjunction with the 2^nd ACM SIGOPS European Conference on Computer Systems (EuroSys) 2007, Lisbon, Portugal, March 20, 2007.

[15] Geoffroy Vallée and Stephen L. Scott. Xen-oscar for cluster virtualization. In ISPA Workshop on XEN in HPC Cluster and Grid Computing Environments (XHPC’06), pages 487–498, December 2006.

[16] Geoffroy Vallée, Thomas Naughton, Hong Ong, and Stephen L. Scott. Checkpoint/restart of virtual machines based on xen. In HAPCW’06: High Availability and Performance Computing Workshop, Santa Fe, New Mexico, USA, October 2006. Held in conjunction with LACSI 2006.

[17] Geoffroy Vallée, Christine Morin, and Stephen L. Scott. A framework for high availability based on a single system image. In HAPCW’05: High Availability and Performance Computing Workshop, Santa Fe, New Mexico, USA, October 2005. Held in conjunction with LACSI 2005.

[18] Geoffroy Vallée, Stephen L. Scott, Christine Morin, Jean-Yves Berthou, and Hugues Prisker. SSI-OSCAR: a cluster distribution for high performance computing using a single system image. In The 3^rd Annual OSCAR Symposium, University of Guelph, Guelph, Ontario, Canada, May 2005. Held in conjunction with the 19th International Symposium on High Performance Computing Systems and Applications (HPCS 2005).

[19] Massimo Coppola, Yvon Jégou, Brian Matthews, Christine Morin, Luis Pablo Prieto, Óscar David Sánchez, Erica Y Yang, Haiyan Yu. "Virtual Organization Support within a Grid-Wide Operating System", IEEE Internet Computing, Vol. 12, No. 2, 2008

[20] C. Morin. XtreemOS: a Grid Operating System Making your Computer Ready for Participating in Virtual Organizations, IEEE International Symposium on Object/component/service-oriented Real-time distributed Computing (ISORC), Santorini Island, Greece, May 2007.

[21] Thomas Ropars. Combining Optimism and Pessimism in a Grid Message Logging Protocol. In Student Forum of International Conference on Dependable Systems and Networks (DSN 2007) (Supplemental Volume), Edinburgh, UK, June 2007

[22] E. Yang, B. Matthews, A. Lakhani, Y. Jégou, C. Morin, O. Sanchez, C. Franke, P. Robinson, A. Hohl, B. Scheuermann, D. Vladusic, H. Yu, A. Qin, R. Lee, E. Focht, M. Coppola. Virtual Organization Management in XtreemOS: an Overview, CoreGrid Symposium, Rennes, France, August 2007.

[23] John Mehnert-Spahn, Michael Schöttner, Thomas Ropars, David Margery, Christine Morin, Julita Corbalán, and Toni Cortes. XtreemOS Grid Checkpointing Architecture. IEEE International Symposium on Cluster Computing and the Grid (poster), Lyon, France, May 19-22, 2008.

[24] Thomas Ropars and Christine Morin. "O2P : un protocole à enregistrement de messages extrêmement optimiste". In Actes de RenPar'18, 2008

[25] Thomas Ropars, Christine Morin. "Fault Tolerance in a Cluster Federation with O2P-CF", In Workshop on Resiliency in High-Performance Computing (Resilience 2008). Held in conjunction with CCGrid 2008

[26] Thomas Ropars, Emmanuel Jeanvoine and Christine Morin. GAMoSe: An Accurate Monitoring Service for Grid Applications. In 6th International Symposium on Parallel and Distributed Computing (ISPDC 2007), Pages 295—302, Hagenberg, Austria, July 2007.

[27] Gallard J., Gallard P., Lebre A., Morin C., Scott S., Vallée G. Refinement Proposal of the Goldberg's Theory. INRIA research report N° RR-6613 (2008)

[28] C. Morin et al. XtreemOS: a Vision for a Grid Operating System, XtreemOS technical report, XosTechRep_04 (http://www.xtreemos.eu), May 2008.

[29] Matthieu Fertré and Christine Morin. Extending a cluster SSI OS for transparently checkpointing message-passing parallel applications. In International Symposium on Parallel Architectures Algorithms, and Networks (I-SPAN05), Las Vegas, Nevada, USA, December 2005.

[30] Adrien Lèbre, Renaud Lottiaux, Erich Focht, and Christine Morin. Reducing kernel development complexity in distributed environments. In Europar 2008, August 2008.

[31] Sébastien Monnet, Christine Morin, and Ramamurthy Badrinath. Hybrid checkpointing for parallel applications in cluster federations. In 4th IEEE/ACM International Symposium on Cluster Computing and the Grid, Chicago, IL, USA, April 2004. CCGrid 2004, IEEE. Electronic version.

[32] John Mehnert-Spahn, Michael Sch¨ottner, and Christine Morin. Checkpointing process groups in a grid environment. In Proc. of the International Conference on Parallel and Distributed Computing (PDCAT ’08), December 2008.

[33] Matthieu Fertré and Christine Morin. Transparent message-passing parallel applications checkpointing in kerrighed. In High Availability and Performance Computing Workshop 2005 (HAPCW05), Santa Fe, New Mexico, USA, October 2005.

[34] Louis Rilling and Christine Morin. Partage de données transparent et tolérant aux fautes pour la grille. In Actes de la 4ème Conférence Fran¸caise sur les Systèmes d’Exploitation (CFSE 4), pages 135–146, Le Croisic, France, April 2005.

[35] Sylvain Jeuland, Yvon Jégou, Oscar David Sanchez, and Christine Morin. Support d’organisations virtuelles au sein d’un système d’exploitation pour la grille. In Actes de RenPar´18, Fribourg, Switzerland, February 2008.

[36] Louis Rilling and Christine Morin. A fault-tolerant transparent data sharing service for the grid. Research report 5427, INRIA, Rennes, France, December 2004.

[37] Emmanuel Jeanvoine, Louis Rilling, Christine Morin, and Daniel Leprince. Using overlay networks to build operating system services for large scale grids. Scalable Computing : Practice and Experience, 8(3) :229–239, September 2007.