Difference between revisions of "Milestones"

From Point
Jump to: navigation, search
(PHASE 1)
Line 29: Line 29:
 
====PHASE 1====
 
====PHASE 1====
 
* Define common multi-core events
 
* Define common multi-core events
* Port KOJAK distributed trace file analysis to cluster environment
+
* Port Scalasca distributed trace file analysis to cluster environment
* Develop additional KOJAK patterns for hardware counter profile data
+
* Develop additional Scalasca patterns for hardware counter profile data
 
* '''Automate use of TAU call-path profiles for selective EPILOG tracing'''
 
* '''Automate use of TAU call-path profiles for selective EPILOG tracing'''
  

Revision as of 20:12, 14 July 2009

Goals that emphasize interaction between institutions are marked in bold.

UO

PHASE 1

  • Add routine-level and message passing AI
  • Develop experimentation framework and user frontend
  • Develop comparative analysis modules
  • Update PerfDMF for regression analysis
  • OTF format updates for EPILOG
  • Evaluate sample-based and direct measurement integration

PHASE 2

  • Develop controller for parametric experiments
  • Develop user-interface for ACA power tool
  • Build PRA prototype, link with software test harness
  • Implement joint measurement support
  • Update PerfDMF for PerfSuite data
  • Incorporate PAPI features for multi-core

PHASE 3

  • Port AI on HPC systems and test
  • Complete AE framework and release
  • Integrate ACA with experimentation system
  • Complete ACA framework and release
  • Integrate PRA with application groups
  • Merge automatic profile analysis with KOJAK

UTK

PHASE 1

  • Define common multi-core events
  • Port Scalasca distributed trace file analysis to cluster environment
  • Develop additional Scalasca patterns for hardware counter profile data
  • Automate use of TAU call-path profiles for selective EPILOG tracing

PHASE 2

  • Implement additional PAPI network components
  • Test distributed trace file analysis on production applications
  • Migrate contextual hardware counter information to PAPI standard
  • Incorporate PerfSuite and TAU profiles into EXPERT analysis

PHASE 3

  • Continue to develop and deploy component PAPI
  • Incorporate events from PAPI components into KOJAK analysis
  • Extend distributed trace analysis to more parallel paradigms
  • Integrate low-overhead statistical hardware counter profiling with TAU and PerfSuite

NCSA

PHASE 1

  • Update core library for current processors/OS
  • Incorporate Perfmon2 in hardware counter library
  • Develop user-oriented reference manual
  • Java API design and development (XML access)
  • Begin integration with PerfDMF
  • Install project software suite at NCSA
  • PerfSuite v1.0

PHASE 2

  • Update core library for current processors/OS
  • Java XML API developed and released
  • Java hardware counter API underway
  • Develop engineering guide
  • Integrate PerfDMF access with PerfSuite tools

PHASE 3

  • Update core library for current processors/OS
  • Integration with PerfDMF completed
  • Joint TAU/PerfSuite automatic analysis tools completed

PSC

PHASE 1

  • Install current project perrformance toolset at PSC
  • Train consultants in the tools' new, advanced analysis modes
  • Assess performance improvement opportunities of NEMO3D, ENZO, and NAMD groups
  • Coordinate AG Performance Engineering seminars: Best Practices

PHASE 2

  • Support performance engineering by NEMO3D, ENZO, NAMD, Cactus groups
  • Apply updated tool features, tracking performance gains
  • Coordinate AG Performance Engineering seminars: Tools for multicore

PHASE 3

  • Inject performance engineering into additional applications
  • Apply updated tool features, tracking performance gains
  • Coordinate AG Performance Engineering seminars: Advanced Tools