Presentation.Text.Marked.Image.Marked-Computer Science-Project Presentation, Slides of Computer Science

Download Presentation.Text.Marked.Image.Marked-Computer Science-Project Presentation and more Slides Computer Science in PDF only on Docsity! Performance Evaluation of Parallel Numerical Libraries docsity.com Outline  Introduction  To Whom it may concern  Parallel Libraries  Underlying Architecture  Cluster  SGI  Performance Metrics  Project Schedule docsity.com To Whom it may Concern  Top500.org provides statistics on parallel computing docsity.com Parallel Libraries  Parallel libraries selected for this project  PBLAS  LAPACK  SCALAPACK  PETSc  Distributed Computing Toolbox (Matlab) docsity.com Netlib.org  Source of FREE mathematical libraries  Collection of mathematical software, papers, and databases  Primary Institutions: – University of Tennessee, Knoxville – Oak Ridge National Laboratory – University of California: Berkeley docsity.com PBLAS  Parallel Basic Linear Algebra Subprograms  Parallel version of BLAS  Performs linear algebra operations on distributed-memory concurrent computers  A major component of the ScaLAPACK library docsity.com PBLAS  Three Levels  Level 1  Vector-Vector Operations (swap, copy, addition, dot product)  Level 2  Matrix-Vector Operations (multiply, rank-updates, outer-product)  Level 3  Matrix-Matrix Operations (multiply, transpose, rank-updates) docsity.com Who Uses PBLAS??  Mainly used in  Scientific and Engineering computations docsity.com Libraries Hierarchy ee PVM/MPL/.. docsity.com PETSc  Portable Extensible Toolkit for Scientific Computation  Very popular parallel library docsity.com Who Uses PETSc??  PETSc has been successfully employed in applications in:  Nano-simulations  Biology/Medical  Imaging and Surgery  Fusion  Environmental/Subsurface Flow  Computational Fluid Dynamics  Wave propagation and the Helmholz equation  Many and many more docsity.com Cluster  Cluster computing distributes the computational load to collections of similar machines  Each Computer (Node) is a complete Computer  Connected to each other via a high speed bus  Job broken up into blocks of data and executed independently  Cheap and easy to produce docsity.com SGI  Super Computer  SGI at PIEAS  64 processors  Connected by a local high-speed computer bus  128 GB memory docsity.com Performance Evaluation Metrics  To analyze the performance of a system, we need some performance evaluation metrics  Speedup  Efficiency  Scalability docsity.com Performance Evaluation Metrics • Scalability – Ability of a system to handle growing amount of work – In parallel computing scalability measures the capability to effectively utilize an increasing number of processors – In most of the cases, scalability decreases as more number of processors are involved in computation docsity.com Challenges in performance evaluation  Many tools for performance data gathering and analysis  PAPI, TAU, SvPablo, Kojak, …  Various interfaces, levels of automation, and approaches to information presentation  User’s point of view - What do the different tools do? Which is most appropriate for a given application? - (How) can multiple tools be used in concert? - I have tons of performance data, now what? - What automatic tuning tools are available, what exactly do they do? - How hard is it to install/learn/use tool ? - Is instrumented code portable? What’s the overhead of instrumentation? How does code evolution affect the performance analysis process? docsity.com Project Work Flow Step 1 • Selection of Parallel Libraries • SCALAPACK, PETSc, PBLAS, ... Step 2 • Selection of Performance Analysis tool • PAPI, TAU, SvPablo, Kojak, … Step 3 • Performance evaluation of algorithms developed at PIEAS • Cross checking of Performance Analyzing Tool & Algorithms Step 4 • Performance Evaluation of Parallel Libraries docsity.com Project Schedule continued  Semester 8  Performance Evaluation of Parallel Numerical Libraries  Thesis Writing docsity.com References  URL:http://top500.org  Alexey Borisenko, “Performance Evaluation in Parallel Systems”, School of information Technology, University of Ottawa, Canada, 2010  URL:http://www.codeproject.com/KB/IP/ClusterC omputing.aspx  URL: http://www.mcs.anl.gov/petsc/petsc- 2/documentation  URL:http://www.netlib.org/scalapack/scalapack_ home.html docsity.com