Browsing by Author "Mandal, Anirban"
Now showing 1 - 5 of 5
Results Per Page
Sort Options
Item Mapping HPF onto the Grid(2003-03-12) Mandal, AnirbanFor this thesis, we have developed a tool for mapping HPF applications onto the Grid using the GrADS infrastructure. To build the mapper, the tool makes use of SPMD task graph representation of the application. Using the mapper generated, we have been able to launch an HPF application, namely tomcatv, on the Grid. To our knowledge this is the only instance of an HPF application running on the Grid. We have compared the generated mapper with the generic site-aware mapper in GrADSoft. The results show that, for Grid runs of the application on 16 processors distributed over three clusters, our mapper has performance comparable to the generic site-aware mapper in GrADS. On the average, the mapper generated by the tool performs about 5% better than the generic site-aware mapper.Item Mapping HPF onto the Grid(2003) Mandal, Anirban; Kennedy, KenFor this thesis, we have developed a tool for mapping HPF applications onto the Grid using the GrADS [ea02] infrastructure. To build the mapper, the tool makes use of SPMD taskgraph representation of the application. Using the mapper generated, we have been able to launch an HPF application, namely tomcatv, on the Grid. To our knowledge this is the only instance of an HPF application running on the Grid. We have compared the generated mapper with the generic site-aware mapper in GrADSoft. The results show that, for Grid runs of the application on 16 processors distributed over three clusters, our mapper has performance comparable to the generic site-aware mapper in GrADS. On the average, the mapper generated by the tool performs about 5% better than the generic site-aware mapper.Item Scalable Grid Application Scheduling via Decoupled Resource Selection and Scheduling(2006-01-20) Casanova, Henri; Chien, Andrew A.; Kee, Yang-Suk; Kennedy, Ken; Koelbel, Charles; Mandal, Anirban; Zhang, YangOver the past years grid infrastructures have been deployed at larger and larger scales, with envisioned deployments incorporating tens of thousands of resources. Therefore, application scheduling algorithms can become unscalable (albeit polynomial) and thus unusable in large-scale environments. One reason for unscalability is that these algorithms perform implicit resource selection. One can achieve better scalability by performing explicit resource selection decoupled from scheduling (aka "decoupled" approach). Furthermore, we hypothesize that one can achieve similar or even better performance as with the non-decoupled approach (aka "one step" approach) by selecting resources judiciously. Leveraging the Virtual Grid Execution System, we demonstrate that the decoupled approach is indeed both scalable and effective in large-scale and highly heterogeneous resource environments.Item Toward a Tool for Scheduling Application Workflows onto Distributed Grid Systems(2006-07-05) Mandal, AnirbanIn this dissertation, we present a design and implementation of a tool for automatic mapping and scheduling of large scientific application workflows onto distributed, heterogeneous Grid environments. The thesis of this work is that plan-ahead, application-independent scheduling of workflow applications based on performance models can reduce the turnaround time for Grid execution of the application, reducing burden of Grid application development. We applied the scheduling strategies successfully to Grid applications from the domains of bio-imaging and astronomy and demonstrated the effectiveness and efficiency of the scheduling approaches. We also proposed and evaluated a novel scheduling heuristic based on a middle-out traversal of the application workflow. A study showed that jobs have to wait in batch queues for a considerable amount of time before they begin execution. Schedulers must consider batch queue waiting times when scheduling Grid applications onto resources with batch queue front ends. Hence, we developed a smart scheduler that considers estimates of batch queue wait times when it constructs schedules for Grid applications. We compared the proposed scheduling techniques with existing dynamic scheduling strategies. An experimental evaluation of this scheduler on data-intensive workflows shows that its approach of planning schedules in advance improves over previous online scheduling approaches. We studied the scalability of the proposed scheduling approaches. To deal with the scale of future Grids consisting of hundreds of thousands of resources, we designed and implemented a novel cluster-level scheduling algorithm, which scales linearly on the number of abstract resource classes. An experimental evaluation using workflows from two applications shows that the cluster-level scheduler achieves good scalability without sacrificing the quality of schedule.Item Toward a tool for scheduling application workflows onto distributed grid systems(2006) Mandal, Anirban; Kennedy, KenIn this dissertation, we present a design and implementation of a tool for automatic mapping and scheduling of large scientific application workflows onto distributed, heterogeneous Grid environments. The thesis of this work is that plan-ahead, application-independent scheduling of workflow applications based on performance models can reduce the turnaround time for Grid execution of the application, reducing burden of Grid application development. We applied the scheduling strategies successfully to Grid applications from the domains of bio-imaging and astronomy and demonstrated the effectiveness and efficiency of the scheduling approaches. We also proposed and evaluated a novel scheduling heuristic based on a middle-out traversal of the application workflow. A study showed that jobs have to wait in batch queues for a considerable amount of time before they begin execution. Schedulers must consider batch queue waiting times when scheduling Grid applications onto resources with batch queue front ends. Hence, we developed a smart scheduler that considers estimates of batch queue wait times when it constructs schedules for Grid applications. We compared the proposed scheduling techniques with existing dynamic scheduling strategies. An experimental evaluation of this scheduler on data-intensive workflows shows that its approach of planning schedules in advance improves over previous online scheduling approaches. We studied the scalability of the proposed scheduling approaches. To deal with the scale of future Grids consisting of hundreds of thousands of resources, we designed and implemented a novel cluster-level scheduling algorithm, which scales linearly on the number of abstract resource classes. An experimental evaluation using workflows from two applications shows that the cluster-level scheduler achieves good scalability without sacrificing the quality of schedule.