Presentation is loading. Please wait.

Presentation is loading. Please wait.

Parallel Tomography Shava Smallen SC99. Shava Smallen SC99AppLeS/NWS-UCSD/UTK What are the Computational Challenges? l Quick turnaround time u Resource.

Similar presentations


Presentation on theme: "Parallel Tomography Shava Smallen SC99. Shava Smallen SC99AppLeS/NWS-UCSD/UTK What are the Computational Challenges? l Quick turnaround time u Resource."— Presentation transcript:

1 Parallel Tomography Shava Smallen SC99

2 Shava Smallen SC99AppLeS/NWS-UCSD/UTK What are the Computational Challenges? l Quick turnaround time u Resource availability and utilization u Network performance u Coallocation l Transparent execution u Single login u Remote data access u Security

3 Shava Smallen SC99AppLeS/NWS-UCSD/UTK GTOMO l Developed by collaboration of NCMIR researchers and computer scientists to address computational challenges of telescience by leveraging distributed resources l GTOMO is an embarrassingly parallel implementation of tomography.

4 GTOMO Description l projections are preprocessed into sinograms l each sinogram is individually processed into a slice sinograms projections slices

5 GTOMO Architecture driver ptomo Off-line Solid lines = data flow dashed lines = control Work queue scheduling reader disk writer disk

6 Shava Smallen SC99AppLeS/NWS-UCSD/UTK Grid Enabled l GTOMO is implemented using components of the Globus toolkit u distributed resources u single login u security l Uses AppLeS to achieve performance u coallocation of workstations and immediately available supercomputer nodes

7 Shava Smallen SC99AppLeS/NWS-UCSD/UTK AppLeS = Application Level Scheduling l AppLeS + application = self-scheduling application l scheduling decisions based on u dynamic information l available from Network Weather Service (NWS) u static application and system information l Methodology u select sets of resources u plan possible schedules for each set of feasible resources u predict the performance for each schedule u implement best predicted schedule on selected infrastructure

8 Shava Smallen SC99AppLeS/NWS-UCSD/UTK AppLeS for GTOMO l Resource selection u NCMIR interactive workstations u NPACI supercomputer time l We have developed a scheduler which coallocates program execution over workstations and immediately available supercomputer nodes for an improved execution performance

9 Shava Smallen SC99AppLeS/NWS-UCSD/UTK Resource Selection l Strategy: u submit GTOMO to available workstations u use dynamic information available from the supercomputer’s batch scheduler to determine a job request which will be started immediately l available on Maui Scheduler l Utilizes computational resources available to a typical research lab

10 Shava Smallen SC99AppLeS/NWS-UCSD/UTK Preliminary Experiment Results l Resources u 6 workstations available at Parallel Computation Laboratory (PCL) at UCSD u immediately available nodes on SDSC SP-2 (128 nodes) l Maui scheduler exports the number of immediately available nodes l e.g. 5 nodes available for the next 30 mins 10 nodes available for the next 10 mins

11 Shava Smallen SC99AppLeS/NWS-UCSD/UTK Allocation Strategies/Experiment Setup l 4 strategies compared: u SP2Immed/WS: workstations and immediately available SP-2 nodes u WS: workstations only u SP2Immed: immediately available SP-2 nodes only u SP2Queue(n): traditional batch queue submit using n nodes l experiments performed in production environment u ran experiments in sets, each set contains all strategies l e.g. SP2Immed, SP2Immed/WS, WS, SP2Queue(8) u within a set, experiments ran back-to-back

12 Experiment Results (8 nodes on SP-2)

13 Experiment Results (16 nodes on SP-2)

14 Experiment Results (32 nodes on SP-2)

15 Shava Smallen SC99AppLeS/NWS-UCSD/UTK Next Steps l Develop contention model to address network overloading which includes u NWS bandwidth measurements u network capacity information l Expansion of platform u reservations (e.g. GARA scheduled resources) u S3 l On-line tomography ( NPACI Telescience Alpha Project)

16 Shava Smallen SC99AppLeS/NWS-UCSD/UTK People l AppLeS: (http://apples.ucsd.edu) u Shava Smallen, Jim Hayes, Fran Berman, Rich Wolski, Walfredo Cirne l NCMIR: (http://www-ncmir.ucsd.edu) u Mark Ellisman, Marty Hadida-Hassan, Jaime Frey l Globus: (http://www.globus.org) u Carl Kesselman, Mei-Hui Su l ssmallen@cs.ucsd.edu


Download ppt "Parallel Tomography Shava Smallen SC99. Shava Smallen SC99AppLeS/NWS-UCSD/UTK What are the Computational Challenges? l Quick turnaround time u Resource."

Similar presentations


Ads by Google