Parallel Tomography

1
Parallel Tomography

• Shava Smallen
• SC99

2
What are the Computational Challenges?

• Quick turnaround time
• Resource availability and utilization
• Network performance
• Coallocation
• Transparent execution
• Single login
• Remote data access
• Security

3
GTOMO

• Developed by collaboration of NCMIR researchers
  and computer scientists to address computational
  challenges of telescience by leveraging
  distributed resources
• GTOMO is an embarrassingly parallel
  implementation of tomography.

4
GTOMO Description

• projections are preprocessed into sinograms

• each sinogram is individually processed into a
  slice

5
GTOMO Architecture
Off-line
Work queue scheduling
Solid lines data flow dashed lines control
6
Grid Enabled

• GTOMO is implemented using components of the
  Globus toolkit
• distributed resources
• single login
• security
• Uses AppLeS to achieve performance
• coallocation of workstations and immediately
  available supercomputer nodes

7
AppLeS Application Level Scheduling

• AppLeS application self-scheduling
  application
• scheduling decisions based on
• dynamic information
• available from Network Weather Service (NWS)
• static application and system information
• Methodology
• select sets of resources
• plan possible schedules for each set of feasible
  resources
• predict the performance for each schedule
• implement best predicted schedule on selected
  infrastructure

8
AppLeS for GTOMO

• Resource selection
• NCMIR interactive workstations
• NPACI supercomputer time
• We have developed a scheduler which coallocates
  program execution over workstations and
  immediately available supercomputer nodes for an
  improved execution performance

9
Resource Selection

• Strategy
• submit GTOMO to available workstations
• use dynamic information available from the
  supercomputers batch scheduler to determine a
  job request which will be started immediately
• available on Maui Scheduler
• Utilizes computational resources available to a
  typical research lab

10
Preliminary Experiment Results

• Resources
• 6 workstations available at Parallel Computation
  Laboratory (PCL) at UCSD
• immediately available nodes on SDSC SP-2 (128
  nodes)
• Maui scheduler exports the number of immediately
  available nodes
• e.g. 5 nodes available for the next 30 mins
• 10 nodes available for the next 10 mins

11
Allocation Strategies/Experiment Setup

• 4 strategies compared
• SP2Immed/WS workstations and immediately
  available SP-2 nodes
• WS workstations only
• SP2Immed immediately available SP-2 nodes only
• SP2Queue(n) traditional batch queue submit using
  n nodes
• experiments performed in production environment
• ran experiments in sets, each set contains all
  strategies
• e.g. SP2Immed, SP2Immed/WS, WS, SP2Queue(8)
• 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
Next Steps

• Develop contention model to address network
  overloading which includes
• NWS bandwidth measurements
• network capacity information
• Expansion of platform
• reservations (e.g. GARA scheduled resources)
• S3
• On-line tomography (NPACI Telescience Alpha
  Project)

16
People

• AppLeS (http//apples.ucsd.edu)
• Shava Smallen, Jim Hayes, Fran Berman, Rich
  Wolski, Walfredo Cirne
• NCMIR (http//www-ncmir.ucsd.edu)
• Mark Ellisman, Marty Hadida-Hassan, Jaime Frey
• Globus (http//www.globus.org)
• Carl Kesselman, Mei-Hui Su
• ssmallen_at_cs.ucsd.edu