[ Bottom of Page | Previous Page | Next Page | Contents | Index ]

IBM Books

Parallel Environment for AIX 5L V4.1
Hitchhiker's Guide


SA22-7947-00

Contents

Figures

About this book

  • Dedication -- Douglas Adams (1952 - 2001)
  • The place for this book in the grand scheme of life, the universe, and everything
  • What is all this?
  • Conventions and terminology used in this book
  • Abbreviated names
  • How to send your comments
  • National Language Support
  • What's new in Parallel Environment 4.1?

    • Hitching a lift on the Vogon constructor ship

  • What is the IBM Parallel Environment for AIX?
  • What is the Parallel Operating Environment?
  • Before you start
  • Running POE
  • Who is in control?

    • The answer is 42

  • Message passing
  • Data decomposition
  • Functional decomposition
  • Duplication versus redundancy
  • Protocols supported
  • Thread debugging implications
  • Checkpointing and restarting a parallel program
  • Limitations
  • How checkpointing and restarting works
  • A checkpoint/restart scenario

    • Don't panic

  • Messages
  • Message catalog errors
  • Finding PE messages
  • Logging POE errors to a file
  • Message format
  • Diagnosing problems using the Install Verification Program
  • Cannot compile a parallel program
  • Cannot start a parallel job
  • Cannot execute a parallel program
  • The program runs but...
  • The parallel debugger is your friend
  • It core dumps
  • No output at all
  • It hangs
  • Let us attach the debugger
  • Other hangups
  • Bad output
  • Debugging and threads

    • So long and thanks for all the fish

  • Tuning the performance of a parallel application
  • How much communication is enough?
  • Tuning the performance of threaded programs
  • Why is this so slow?
  • Profile it
  • Parallelize it
  • Wrong answer!
  • Here is the fix!
  • It is still not fast enough!
  • Tuning summary

    • Babel fish

  • Point-to-point communication
  • SEND (Non-Blocking)
  • RECEIVE (Non-Blocking)
  • SEND (Blocking)
  • RECEIVE (Blocking)
  • SEND/RECEIVE (Blocking)
  • STATUS
  • WAIT
  • TASK_SET
  • TASK_QUERY
  • ENVIRON
  • STOPALL
  • PACK
  • UNPACK
  • VSEND (Blocking)
  • VRECV (Blocking)
  • PROBE
  • Collective communications
  • BROADCAST
  • COMBINE
  • CONCAT
  • GATHER
  • INDEX
  • PREFIX
  • REDUCE
  • SCATTER
  • SHIFT
  • SYNC
  • GETLABEL
  • GETMEMBERS
  • GETRANK
  • GETSIZE
  • GETTASKID
  • GROUP
  • PARTITION
  • Reduction functions
  • User-defined reduction functions
  • Global variables and constants
  • Last error code
  • Wildcards
  • General notes
  • Task identifiers
  • Message length
  • Creating MPI objects
  • Using wildcard receives
  • Reduction functions
  • Error handling
  • Before and after using MPI functions
  • Using message passing handlers

    • Mostly harmless

  • What is a safe program?
  • Safety and threaded programs
  • Using threaded programs with non-thread-safe libraries
  • Message ordering
  • Program progress when two processes initiate two matching sends and receives
  • Communication fairness
  • Resource limitations

    • Appendix A. A sample program to illustrate messages

  • Figuring out what all of this means

    • Appendix B. Parallel Environment internals

  • What happens when I compile my applications?
  • How do my applications start?
  • How does POE talk to the nodes?
  • How are signals handled?
  • What happens when my application ends?

    • Notices

  • Trademarks
  • Acknowledgements

    • Glossary

    Bibliography

  • Information formats and them on the Web
  • System planning publications
  • RS/6000 SP Switch Router publications
  • Related hardware publications
  • RS/6000 SP software publications
  • AIX publications
  • Reliable Scalable Cluster Technology (RSCT) publications
  • Cluster Systems Management (CSM) publications
  • DCE publications
  • Redbooks
  • Non-IBM publications

    • Index

    [ Top of Page | Previous Page | Next Page | Contents | Index ]