The Performance of PVM
on MPP Systems

Henri Casanova Jack Dongarra Weicheng Jiang

Thu Jul 20 07:23:10 EDT 1995

Abstract:

• Introduction
• The Semantics of PVM Message Passing
  • Terminology
    • Synchronous Send
    • Asynchronous Send
    • Blocking Send
    • Non-blocking Send
    • Blocking Receive
    • Non-blocking Receive
  • The Communication Model
  • The Message-Passing Functions in PVM
    • pvm_send()-pvm_recv-pvm_nrecv()
    • pvm_psend()-pvm_precv()
  • Summary
  • Implementation on MPPs
• The CM-5
  • The Native Message-Passing System
    • CMMD_send_block()-CMMD_receive_block()
    • CMMD_send_async()-CMMD_receive_async()
    • CMMD_send_noblock()
  • Comparison of the Native Routines
  • The Bandwidth
  • The Latency
• The Intel Paragon
  • The Native Message-Passing System
    • isend -irecv
    • csend()-crecv()
  • Comparison of the Native Routines
  • The Bandwidth
  • The Latency
• The IBM SP-2
  • The Native Message-Passing System
    • mpc_bsend()-mpc_brecv()
    • mpc_send()-mpc_recv()
  • The Bandwidth
  • The Latency
• Summary
• Conclusion
• Details on the tests
  • Test program
  • CM-5
  • Intel Paragon
  • IBM SP2
• References
• References
• About this document ...