Table of Contents
T1. Advanced mathematical techniques for LCG
Task coordinator : A. Kryukov, belonging to team: SINP-MSU
Objectives
The main goal of the task is development and implementation of advance algorithms for job scheduling and resource allocation. The investigation of behavior of the algorithms on the real LCG infrastructure permits to evaluate the efficiency of the algorithms and improve the quality of LCG infrastructure.
Methodology
Statistical investigation of monitoring information about the site resources. Use of elements of queuing theory.
Task Input:
The task is depending on:
Deployment of LCG for creating Russian GRID segment Experience in research and development of Recourse Broker in EDG project and tools for submitting application jobs. Experience in deploying EDG middleware in Russian GRID segment.
Result, milestones :
Milestones, results expected and deliverables are defined in the subtasks.
T1.1. Advanced algorithms and tools for resource allocation and job scheduling
Subtask coordinator : M. Sgaravatto, belonging to team: INFN-Padova, deputy A. Kryukov (SINP-MSU)
Other participants: V. Edneral, L. Shamardin and A. Demichev (all SINP-MSU), U. Gasparini and M. Michelotto (INFN-Padova)
Objectives
One of the major components of LCG middleware is the Resource Broker (RB). It is responsible for job schedulingand resource allocation. Thus this service is very important for effective the LCG functioning. In current EDG (and therefore LCG-1) realization, rather straightforward algorithm is implemented: jobs are sent to resources which matches the job requirements, and which are ò??closeò?? to the needed data. We propose to study and implement more sophisticated algorithms, allowing better performance of the overall system for the considered set of applications. In particular we propose to analyze and exploit in the Resource Broker statistical investigation of monitoring information about resources. In the same time we are going to test proposed algorithms on real LCG infrastructure, that allows to improve both the algorithms and RB and, as a result, the quality of GRID services.
Methodology
Statistical investigation of monitoring information about the site resources. Use of elements of queuing theory and use of simulation tools, e.g. MONARC.
Task Input:
Experience in research and development of RB in EDG project and tools for submitting application jobs (in CMS). Experience in deploying EDG middleware in Russian GRID segment.
Result, milestones:
MILESTONES:
- developing of queuing based algorithm for RB, Feb. 2004 - July 2004;
- programming implementation in RB of this algorithm, Sept. 2004 - March 2005;
- testing in LCG, Apr. 2005 - Dec. 2005.
RESULTS EXPECTED:
Program for job scheduling in LCG.
DELIVERABLES:
Preprints, reports, publications in journals. Computer programs and the Program reference manual.
T1.2 Task Title : Large scale data flow simulation in local and GRID environment
Subtask coordinator : V. Kolosov, belonging to team: ITEP
Other participants: S. Makarychev, I. Korolko, E. Lyublev and N. Tsvetkov (all ITEP)
Objectives :
Simulation of large scale data flows in the LCG infrastructure, including LAN and GRID aspects, to reveal weaknesses and bottlenecks as concern behaviour of storage servers under critical loads. Work out recommendations for improvement of the operability.
Methodology :
Use of existing (e.g., MONARC) simulation tools and tools under development. Study of data flows to be produced by real applications of LHC experiments.
Task Input:
Experience in scripting Shell and Perl languages, and in using SNMP monitoring toolkits. Experience in design and development of large applications in GRID environment.
Result, milestones :
MILESTONES:
- Study of requirements on the storage device loading under real applications. Feb. 2004 - Apr. 2004;
- Design and development of a preliminary version of the simulation tools, May 2004 - Sep. 2004;
- Simulating a storage server behaviour under critical loads, Nov. 2004 - Nov. 2005;
RESULTS EXPECTED:
Analytic reports, regarding behavior of LCG under moderate and heavy load. New tools for GRID simulation.
DELIVERABLES:
Reports on conferences and publications in preprints and journals.