论文信息 - Using the eNANOS Low-Level Support in the GRMS Framework

Using the eNANOS Low-Level Support in the GRMS Framework

AbstractThe eNANOS is an execution framework developed in the Barcelona Supercomputing Center. One of its mainobjectives is to provide a framework to execute multilevel parallel applications with low-level support. It is also ableto provide information about the execution behavior of applications in run time. This information can be used by aGrid Resource Broker or metascheduler to improve its scheduling and resource strategies and the execution platformcan improve the execution time of applications and resource usage as well. In this paper we discus the steps that wehave to follow to integrate the eNANOS execution environment into the GRMS infrastructure, developed by PSNC.In particular we are interested in the mechanisms to allow the integration of the different components and how to usethe information provided by eNANOS to improve the scheduling strategies in the GRMS system. 1 Introduction The eNANOS is an execution framework developed in the Barcelona Supercomputing Center. One of its main ob-jectives is to provide a framework to execute multilevel parallel applications with low-level support. Furthermore theeNANOS architecture is based on the idea of coordination between the different layers [9]. Currently the eNANOSExecution platforms uses the eNANOS Grid Resource Broker [8] which manages the jobs from the Grid layers incoordination with the local resource environments.The eNANOS System is also able to provide information about the execution behavior of applications in run time,such as its progress or the obtained performance in a given moment. This information can be used by a Grid ResourceBroker or metascheduler to improve its scheduling and resource strategies and the execution platform can improve theexecution time of applications and resource usage as well.The main effort of the PSNC in the Grid resource management is the GRMS resource broker [2]. GRMS is anopen source meta-scheduling system for large scale distributed computing infrastructures. Based on the dynamicresource selection, mapping and advanced grid scheduling methodologies, it has been tailored to deal with resourcemanagement challenges in Grid environments, e.g. load-balancing among clusters, setting up execution environmentsbefore and after job execution, remote job submission and control, ﬁles staging, workﬂow management and more.

[1] Raj Vaswani,et al. A dynamic processor allocation policy for multiprogrammed shared-memory multiprocessors , 1993, TOCS.

[2] Jesús Labarta,et al. Multi-Criteria Grid Resource Management Using Performance Prediction Techniques , 2007 .

[3] Alejandro Duran,et al. Dynamic load balancing of MPI+OpenMP applications , 2004 .

[4] Jesús Labarta,et al. eNANOS: Coordinated Scheduling in Grid Environments , 2005, PARCO.

[5] Jarek Nabrzyski,et al. Multicriteria aspects of Grid resource management , 2004 .

[6] Jarek Nabrzyski,et al. Dynamic grid scheduling with job migration and rescheduling in the GridLab resource management system , 2004, Sci. Program..

[7] Alejandro Duran,et al. Dynamic load balancing of MPI+OpenMP applications , 2004, International Conference on Parallel Processing, 2004. ICPP 2004..

[8] Jesús Labarta,et al. Uniform Job Monitoring using the HPC-Europa Single Point of Access , 2006, Sixth IEEE International Symposium on Cluster Computing and the Grid (CCGRID'06).

[9] Jesús Labarta,et al. How the JSDL can exploit the parallelism? , 2006, Sixth IEEE International Symposium on Cluster Computing and the Grid (CCGRID'06).