160 likes | 248 Views
ProActive Task Manager Component for SEGL Parameter Sweeping. Natalia Currle-Linde and Wasseim Alzouabi High Performance Computing Center Stuttgart (HLRS), University of Stuttgart http://www.coregrid.net.
E N D
ProActive Task Manager Component for SEGL Parameter Sweeping Natalia Currle-Linde and Wasseim Alzouabi High Performance Computing Center Stuttgart (HLRS), University of Stuttgart http://www.coregrid.net
SEGL is a problem solving environment which allows end-user programming of complex, computation-intensive simulation and modeling experiments for science and engineering. • SEGL enables the automated creation, starting and monitoring of complex experiments and supports their effective execution on the GRID. The user of SEGL does not need to have a knowledge of a specific programming language and the GRID structure. SEGL allows the description of complex experiments using the Grid Concurrent Language (GriCoL). • SEGL is a tool for generating parameter sets automatically, issuing jobs in the Grid environment, controlling the successful operation and termination of these jobs, collecting results, and generating new parameter sets based on previous results in order to approach a functional optimum. European Research Network on Foundations, Software Infrastructures and Applications for large scale distributed, GRID and Peer-to-Peer Technologies
J2EE • JBOSS Application Server • JDO (OODB) European Research Network on Foundations, Software Infrastructures and Applications for large scale distributed, GRID and Peer-to-Peer Technologies
Philosophy of GriCoL • Set of components within the tasks is very limited • (e.g gradient searches, genetic algorithms). • Components instandard form, can be repeatedly • reused for modeling complex systems. • If one of the desired components is not available, it is much simpler to implement this component and add it to the language than to generate a new complex application. • Solution: a component based language for describing complex modeling experiments with a sufficient level of abstraction so that • Scientist doesn`t require • knowledge of Grid . • knowledge of parallel programming. European Research Network on Foundations, Software Infrastructures and Applications for large scale distributed, GRID and Peer-to-Peer Technologies
Common properties and principle of organization • Universal language • Graphical-based lanquage – is based on component-structure model • Parallel nature • Parallel processing of many data sets at all levels • Possibility of parallel executions unlimited • asynchronous pipeline operations • Based on principle of wrapping functionality in components • Extensibility of language • Multilayer language • Control flow • Data flow European Research Network on Foundations, Software Infrastructures and Applications for large scale distributed, GRID and Peer-to-Peer Technologies
European Research Network on Foundations, Software Infrastructures and Applications for large scale distributed, GRID and Peer-to-Peer Technologies
Control Flow • User defines the sequence of execution of experiment blocks • Solver block simple parameter sweep • Conrol block program object: allows changing sequence of execution operation according to specified criteria • Connection Lines • blue (asynchronous pipeline operations) • after completion of a program run within • a block, control is transferred to the next • block • red • control is not transfered before all runs in the • previous block have been finished European Research Network on Foundations, Software Infrastructures and Applications for large scale distributed, GRID and Peer-to-Peer Technologies
Data Flow • Manipulation of data in a very fine - grained way • Solver Block: • computation module C • replacemant module R • parameterization module P • data base Each module: Java object • has standard structure • consists of several sections Computation module : • organizes preparation of input data • generates jobs • initializes/controls record of results in DB • controls execution of module operation i European Research Network on Foundations, Software Infrastructures and Applications for large scale distributed, GRID and Peer-to-Peer Technologies
Data Flow (variants of parameterization) European Research Network on Foundations, Software Infrastructures and Applications for large scale distributed, GRID and Peer-to-Peer Technologies
European Research Network on Foundations, Software Infrastructures and Applications for large scale distributed, GRID and Peer-to-Peer Technologies
Scheduler • Administer deployment and maintenance of a list of jobs over the Grid. • Launched on a grid and controlled by a command line interface to execute jobs (complete applications). • Embedded directly in the application and called via the API to execute tasks. • Built on top of ProActive deployement framework. • Result of a collaboration between 3 active objects (Scheduler,JobManager,and ResourceManager) European Research Network on Foundations, Software Infrastructures and Applications for large scale distributed, GRID and Peer-to-Peer Technologies
European Research Network on Foundations, Software Infrastructures and Applications for large scale distributed, GRID and Peer-to-Peer Technologies
Scheduler: main object /non GUI daemon that receives /parses the Jobs. • Job manager: Management of Submitted Jobs. Has job description, deployment and policies. • Resource Manager: resource allocation and management. European Research Network on Foundations, Software Infrastructures and Applications for large scale distributed, GRID and Peer-to-Peer Technologies
European Research Network on Foundations, Software Infrastructures and Applications for large scale distributed, GRID and Peer-to-Peer Technologies
European Research Network on Foundations, Software Infrastructures and Applications for large scale distributed, GRID and Peer-to-Peer Technologies
European Research Network on Foundations, Software Infrastructures and Applications for large scale distributed, GRID and Peer-to-Peer Technologies