Description
Cloud Computing has emerged as a very promising paradigm to simplify and improve the management of current IT infrastructures of any kind. Clouds have opened up avenues in this area to ease the maintenance, operation and use of Grid sites, and to explore new resource sharing models that could simplify in some cases the porting and development of Grid applications. In the first part of the session the selected talks will provide an overview of novel cloud provisioning models and virtualization technologies for Grid infrastructures. Then the second part will provide an overview of real-life experiences based on the multi national cloud infrastructure BalticCloud. The BalticCloud is a cross national (EST, LT, LV, BY) cloud infrastructure based/experimenting with many open source solutions (Eucalyptus, Enomaly, OpenNEbula, and more), and with grid users used to gLite, ARC and UNICORE. In addition the BalticCloud is connecting to the new Northern Europe Cloud (SE, NO, DK, Poland and more).
Agenda:
- "Enabling Distributed Job Submission in Dynamic Virtual Execution Environments for EGEE Users", D.Salomoni, M.Cecchi, A.Ghiselli, A.Italiano, M.Orrรน, D.Rebatto, V.Venturi, L.Zangrando (INFN)
Abstract: Grid infrastructures are often limited in terms of execution environments provided to users, e.g. for what regards architecture, operating system, or the presence (or absence) of certain libraries. At INFN, we have developed a software called Worker Nodes on Demand (WNoD), where compute (worker) nodes are dynamically created on virtual machines for the sole purpose of executing a user job. The images of these virtual machine can be different for different users, thus providing support for customized execution environments. WNoD is under progressive deployment at the INFN CNAF Tier-1 center, and it has been been integrated with gLite submission tools. This allows EGEE users to select their virtual image of choice for a job through standard Job Description Language (JDL) statements. WNoD is also being extended to provide access to cloud-like services either locally or via external providers (to expand computing capacity).
- "Virtual Machines at a Tier-1 site." Sander Klous (Nikhef)
Amsterdam hosts one of the Tier-1 sites for the LHC experiments. This particular Tier-1 is part of a larger e-science infrastructure in the Netherlands, called 'Big Grid'. Recently we have started a working group to investigate the possibility to deploy various kinds of Virtual Machines on the Big Grid infrastructure. In my presentation I will address the motivation, the status and the challenges of this project.
- "Experiences with Virtualization of Grid Infrastructure", David O'Callaghan and Brian Coghlan. (TCD)
TCD and Grid-Ireland have been deploying grid systems on virtual machines for five years in Irish and international grid infrastructures. Grid-Ireland currently manages more than ten sites on VMs and operates VM testbeds for development, porting, certification and training. Recent work includes setting up redundant VM hosts with transparent replication for national or central grid services.
- "The batch virtualization project at CERN"., Sebastien Goasguen, Ewan Roche, Tony Cass and Schwickerath Ulrich
Between March and August 2009 a project has been set up at CERN with the aim to evaluate possibilities to use virtualization at a large scale, with the focus on batch computing. Two key issues have been identified for this specific application: the placing of virtual machines on an appropriate hyper-visor, and the selection of an appropriate image which should be driven by the actual demand.
Both commercial and free software solutions exist which are able to solve the placing issue. The virtual machine orchestrater, VMO, a commercial solution by Platform computing, and the free software solution OpenNebula have been evaluated during the project. For VMO, the vendor provided a first implementation of an algorithm for selecting the image to be deployed, which is driven by user requirements of pending batch jobs. For OpenNebula, an external mechanism needs to be developed to perform this task. In the presentation, the basic concepts of the project and lessons learned will be presented. Further visions and possible implications for services offered at CERN will be described.
The session concludes with a panel where relevant researchers of the Grid and Cloud arena will debate about these two technologies, current trends and opportunities for the European Grid Infrastructure. The following speakers will share his view with the audience:
- Ignacio M. Llorente (UCM, RESERVOIR Project)
- Cal Loomis (CNRS, StratusLab Initiative)
- Aake Edlund (KTH, Northern Europe Cloud)
- Sander Klous (NIKHEF, EGEE)
- Ulrich Schwickerath (CERN, EGEE)
- Ilja Livenson (kbfi ,Baltic Cloud)
