Table of ContentsAn Insiders Guide: Janus Optimization and Troubleshooting Overview Sigportals - Message Passing Failures Sigportal Problems by Number More Comments on Sigportals Overview - the Debugger What to do when large production jobs stop making progress and you'd like to figure out why (or gather enough information to beat on your code team, at least.) Other Useful Debug Commands After Attaching Debug’s ‘where’ Command Bug Debug: Using Watchpoints Debug : Using Watchpoints Example Debug: Compiling and Running the Watchpoint Example Overview - NQS What NQS Topics We Can Cover Grabnodes.ksh -- Your Personal Interactive Partition Using NQS Warnlimit for an Orderly Shutdown before Time Expires Example: Using NQS Warnlimit, Sample Program Example: Using NQS Warnlimit, Compile Line and qsub Script Example: Using NQS Warnlimit, Output and Results Overview - OpenMP OMP Introduction OMP Usage/Compiling/Linking OMP Refs. & Directives http://www.openmp.org PGI Users Guide: http://www.pgroup.com/ppro_docs/pgiws_ug/pgi30u.htm OMP C Pragmas OMP Clauses OMP Sample Program, Compile and Run OMP Need-to-do’s! Virtual Nodes Overview - the Profiler Profiler Introduction Profiler Overview Profiler Features Cpu time/HW counters Profiler Features Memory usage/Env. variables Profiler Features More Env. Variables Profiler Features Call Tree/Ancestor Tree/Load Bal. Profiler: Compiling/Running/Getting Results Profiler example exam1.c - the basics on 1 node Profiler example exam1.c - Profile output Profiler Example exam1.c - Call Tree Profiler Example exam1.c - Running on 2 nodes Profiler Example exam1.c - Call Tree for 2 nodes Profiler Example exam2.c - Using the HW Counters to Gets Flops by Routine Profiler Example exam2.c - Display the flops Profiler & Load Balance Sample Code First Look with Profiler, Basic and PROFILE_GSYNC_ROUTINE Looking at wait1() & wait2() Interpreting the Output Profiler Notes - 2nd CPU and Useful HW Counters Manual Profiling - genperf and perfrep Overview - CTH Profiler Example Another Profile Example -- CTH PPT Slide Taking the next step… More OMP CTH w/ “full” OMP utilization Overview - IO Cliff Notes Optimizing IO Rules for best IO performance IO Bottleneck - fyods Rules for Stream IO and Shared File IO File systems - NFS (slowest) & UFS (still slow) File systems PFS file system C and Fortran IO APIs IO Results Impacted by Many Factors IO Results - C Streams IO Results - C write/read/cwrite/cread IO Results - C Shared File - Without ‘yod -masync’ IO Results: C Shared File with ‘yod -masync’ and Best API IO Results: F77 read/write IO Results: F90 using Async IO Sample programs using sync IO and async IO Overview - Math Libs ASCI Red Math Libraries What Math Libraries Exist LIBCSMATH Libcsmath R2.8 enhancements LIBWC LAPACK, PBLAS, ScaLAPACK A new Optimization Tool Example |
Authors: Ben Cole, Greg Henry and Pat Fay
Email: pfay@lanl.gov |