AppLeS. http://www-cse.ucsd.edu/groups/hpcl/apples/apples.html.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....to achieve a high absolute speed on an expensive vector processor such as a Cray T90. It is up to the Grid software to decide which of these resources should best be assigned to execute the application. Resource Cost vs. Performance. As part of the Application Level Scheduler (AppLeS) project [2, 5, 32], we have constructed scheduling agents that make allocation decisions based strictly on deliverable performance. The AppLeS scheduling agent evaluates each resource that is available by considering how quickly it can execute the application (measured in wall clock time) That is, the agent uses ....

AppLeS. http://www-cse.ucsd.edu/groups/hpcl/apples/apples.html.

....will be for a given time frame. We think of this functionality as being analogous to weather forecasting, and as such, term the service the Network Weather Service (NWS) We have developed the NWS for use by schedulers in a networked computational environment. The AppLeS scheduling methodology [4, 2] makes extensive use of its facilities and we are currently implementing versions for Legion [17, 25] and Globus Nexus [10, 15] Initial scheduling results using the NWS are promising [5] In this paper, we focus on the problem of network performance forecasting within the context of scheduling, ....

....25] metacomputing systems. These versions will be initially deployed as part of the GUSTO (Globus UbiquitouS Testbed) 10] and DOCT (Distributed Object Computational Testbed) 1] metacomputing testbeds. We plan to use these implementations both to investigate metacomputing scheduling via AppLeS [4, 2] and the development of general quality of service mechanisms. Acknowledgements The NWS is part of the AppLeS project, and as such, owes much of its existence to Francine Berman and the members of the AppLeS Corps at UCSD whom we thank for pearls of wisdom too plentiful to enumerate. We would ....

AppLeS. http://www-cse.ucsd.edu/groups/hpcl/apples/apples.html.

....the partitioning problem optimally is NP complete, so it is necessary for the user to employ heuristics to arrive at a good solution. 2. 1 Deriving Partitions that Optimize Resource Performance The version of Jacobi2D we use in this example is written in a data parallel SPMD style using KeLP [6, 5]. The KeLP system provides high level abstractions, in the form of C objects, that support runtime data decomposition. In addition, the details associated with message passing in distributed memory computing environments are buried in the abstrac4 P 0 P 1 P 2 Figure 2: Strip data ....

Fink, S., http://www-cse.ucsd.edu/groups/hpcl/scg/kelp.html, 1995.

....we have developed separate implementations of the NWS for the Globus Nexus [8] and Legion [11] metacomputing environments, each of which provides a software infrastructure that supports highperformance distributed and parallel computing. As part of the AppLeS (Application Level Schedulers) project [2, 1] we are developing scheduling agents that make decisions based on applicationlevel performance estimates. The NWS functionality is motivated by the requirements of these agents. In addition, quality of service guarantees in shared network environments (e.g. The Internet) are difficult to achieve. ....

AppLeS. http://www-cse.ucsd.edu/groups/hpcl/apples/apples.html.

....metacomputing systems. These versions will be initially deployed as part of the GUSTO (Globus UbiquitouS Testbed) 10] and DOCT (Distributed Object Computational Testbed) 1] metacomputing testbeds. We plan to use these implementations both to investigate metacomputing scheduling via AppLeS [4, 2] and the development of general quality of service mechanisms. Acknowledgements We would like to thank Francine Berman, Chandra Krintz, and Jenny Schopf at UCSD and Allen Downey at UCB and SDSC for their comments, suggestions, and criticisms, all of which were invaluable. ....

AppLeS. http://www-cse.ucsd.edu/groups/hpcl/apples/apples.html.

....using sockets (based on the netperf [15] utility) and for the Globus Nexus [8] and Legion [13] metacomputing environments, each of which provides a software infrastructure that supports high performance distributed and parallel computing. As part of the AppLeS (ApplicationLevel Schedulers) project [2, 1] we are developing scheduling agents that make decisions based on application level performance estimates. The functionality of the NWS is motivated by the requirements of these agents. In addition, quality of service guarantees in shared network environments (e.g. The Internet) are difficult to ....

AppLeS. http://www-cse.ucsd.edu/groups/hpcl/apples/apples.html.

....time frame in which the application will be scheduled. The Heterogeneous Application Template (HAT) is an interface in which the user provides specific information about the structure, characteristics and current implementations of the 2 For more information on AppLeS, visit the AppLeS home page [1]. ACTUATOR RESOURCE SELECTOR PLANNER PERFORMANCE ESTIMATOR HETEROGENEOUS APPLICATION TEMPLATE USER SPECIFICATIONS INFORMATION POOL RESOURCE MANAGEMENT SYSTEM NETWORK WEATHER SERVICE COORDINATOR MODELS Figure 1. Organization of an AppLeS agent. application and its tasks. The Models are used for ....

APPLES. http://www-cse.ucsd.edu/users/berman/ apples.html.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC