Overview of reports

1
Overview of reports

• Adam Leko
• HCS Research Laboratory
• University of Florida

2
Programming report overview (report form)

• Report on programming
• CAMEL
• Description of benchmark
• Overview of code
• Comparison of UPC code with MPI
• Bench9
• Description of benchmark
• Overview of code
• Comparison of UPC code with MPI and SHMEM
• Discussion of the effect of different
  optimizations on code
• Convolution
• Description of benchmark
• Overview of code
• Comparison of UPC code with MPI and SHMEM
• Discussion of the effect of different
  optimizations on code

3
Report on Paradyn (slideset)

• Overview of Paradyn
• Description of instrumentation overview
• Paradyn architecture overview
• Dynamic instrumentation overview
• Description of metrics and visualizations
• PCL MDL configuration files
• Overview of W3 search model (why, where, when)
  performance consulatant
• Continuation of description based on site visit
  slides
• Demo
• Show performance consultant/W3 on sequential and
  MPI program
• Evaluation of Paradyn discussion
• Finish evaluation from the site visit slides
• Current limitations/caveats
• What problems we might encounter if we try to
  adapt for SHMEM and UPC

4
Report on performance modeling

• Paper outline
• Abstract Introduction
• Overview of performance models
• Reasons for examining performance models in the
  design of a PAT
• Get a list of performance factors
• Performance models can guide a PAT in giving
  advice or identifying bottlenecks
• Overview analysis of current performance models
• Formal performance models
• Petri nets process algebras
• Quick description not a detailed discussion
  since formal performance models arent very
  useful for a PAT
• Analytical performance models
• Meant to give programmer mental picture of
  programming environment (LogP variants, BSP,
  etc)
• Predictive performance models
• Meant to predict performance of actual
  applications on a range of architectures
• Probably the most useful to us
• For each performance model,
• Give list of performance factors used to
  characterize model
• Report accuracy for each model
• Conclusion recommendations

5
Performance model papers

• Formal models
• Software Performance Modeling using PEPA nets,
  S. Gilmore et all.
• Analytical models
• Parallelism in Random Access Machines, S.
  Fortune and J. Wyllie.
• LogP Towards a Realistic Model of Parallel
  Computation, Culler et all.
• A Bridging Model for Parallel Computation, L.
  Valiant.
• Related paper Using the BSP Cost Model for
  Optimal Parallel Neural Network Training, R.
  Rogers and D. Skillicorn.
• Related paper Theory, practice, and a tool for
  BSP performance prediction, J. Hill et all.
• Modeling Performance of Parallel Programs, W.
  Meira, Jr. tech report on lost cycles analysis
• Related paper Parallel Performance Prediction
  Using Lost Cycles Analysis, M. Crovella and T.
  LeBlanc.
• Predictive models
• A Static Parameter based Performance Prediction
  Tool for Parallel Programs, T. Fahringer and H.
  Zima.
• A Framework for Performance Modeling and
  Prection, A. Snaveley et all.
• Predicting and Evaluating Distributed
  Communication Performance, K. Cameron and R. Ge.
• PerPreT A Performance Prediction Tool for
  Massively Parallel Systems, J. Brehm.
• PACE A Toolset to Investigate and Predict
  Performance in Parallel Systems, D. Kerbyson et
  all.
• Related paper Is Predictive Tracing Too Late
  for HPC Users? D. Kerbyson et all.
• An Effective and Practical Performance
  Prediction Model for Parallel Computing on
  Non-dedicated Heterogeneous NOW, Y. Yan and X.
  Zhang.
• Parametric Micro-level Performance Models for
  Parallel Computing, Y. Kim et all.

6
Performance model papers

• Predictive models, continued
• Compiling Performance Models from Parallel
  Programs, A. van Gemund.
• Related paper Performance Prediction of
  Parallel Processing Systems The PAMELA
  Methodology, A. van Gemund.
• Analytical Performance Prediction on
  Multicomputers, M. Clement.
• Studying the Performance Properties of
  Concurrent Programs by Simulation Experiments on
  Synthetic Programs, R. Candlin.
• Performance Prediction of Large Parallel
  Applications Using Parallel Simualtions, R.
  Bagrodia et all.
• Predictive Performance and Scalability Modeling
  of a Large-Scale Application, D. Kerbyson et
  all.
• Parallel Program Performance Prediction Using
  Deterministic Task Graph Analysis, V. Adve.
• Related paper Analyzing the Behavior and
  Performance of Parallel Programs, V. Adve.
• Execution-Driven Tools for Parallel Simulation
  of Parallel Architectures and Applications, D.
  Poulsen and R. Yew.
• Analytical Performance Prediction on
  Multicomputers, M. Clement and M. Quinn.
• A Performance Evaluation of Cluster-based
  Architectures, X. Qin and J. Baer.
• A Source Code Analyzer for Performance
  Prediction, M. Kuhnemann et all.
• Interpretive Performance Prediction for High
  Performance Application Development, M. Parashar
  and S. Hariri.
• Related paper Compile-Time Performance
  Prediction of HPF/Fortran 90D, M. Parashar and
  S. Hariri.
• Performance Prediction A Case Study Using a
  Scalable Shared-Virtual-Memory Machine, X. Sun
  and J. Zhu.
• Accurate Performance Prediction for Massively
  Parallel Systems and its Applications, J. Simon
  and J. Wierum.
• A Performance Modeling System for
  Message-Passing Parallel Programs, D. Grove and
  P. Coddington.
• Related paper Performance Modeling of
  Message-Passing Parallel Programs, D. Grove (PhD
  thesis).

7
Report on compiler optimizations

• Paper outline
• Abstract Introduction
• Overview of compiler optimizations
• Reasons for examining compiler optimizations
• Get a list of performance factors
• Compiler optimizations can affect performance in
  positive or negative ways, so its important to
• Overview analysis of compiler optimizations
• Concentrate on compiler optimizations directed at
  parallel programs (High-Performance Fortran)
• Conclusion recommendations
• Give recommendation on how to best use a
  performance model in the context of our PAT
• Future work (next semester)
• Examine sequential compiler optimizations
• Examine algorithm optimization techniques