Cplant Release 0.3 Notes

August 19, 1999

This release of Cplant software enables independent parallel I/O and contains several bug fixes that increase the reliability of the system and help identify failing components. Users can now access independent filesystems residing on different RAIDS. Access to each RAID is provided by fyod, a daemon process running on the I/O nodes. Application programs can expect to see 15-20 MB/s I/O bandwidth per individual RAID. For more information on using the parallel I/O capabilities, look here.

Highlights

• Independent parallel I/O capability
• Increased message passing reliability
• New showmesh_tk tool for system management
• Backtrace available from hung processes
• Suppression of Fortran runtime library error messages

Bug Fixes

• Unreliable message passing
  • MPI communication errors (hung nodes)
  • Application load errors (unsuccessful loads, hung nodes)
• Yod node list processing
• Compile scripts allow quoted arguments

System Software Developer Tests

• NAS Parallel Benchmarks version 2.3 Class A up to 256 nodes
• MPLinpack benchmark on 310 nodes
• All to All bit error tests on 310 nodes
• Application tests on 256 nodes

Application Tests

• MPI Communication tests
  • Parallel "Hello, world"
  • Ping pong
  • Repeated broadcast
  • Sparse matrix-vector multiply
  • Molecular dynamics exchange
• Application libraries
  • BLAS
  • LAPACK
  • SCALAPACK
  • Aztec
• Applications
  • MP-Quest
  • Tramonto

Changes Since Release 0.2

• fyod support
• showmesh_tk tool added
• fix to the Portals module for message state handling
• fix to the compile scripts to allow quoted arguments
• fix to yod for processing node list
• fixed problem with GDB being parent of application process
• PCT cleans up after child process rather than using atexit()
• 'yod -g' now gives backtrace when interrupted by the user
• suppression of FORTRAN runtime library error messages (yod -d rtl)
• more group communication failure info
• more verbose launch failure messages
• yod sends mail on failed loads
• server communication library Portals now marked as allocated
• several updates to the regression test scripts
• new userlog entries flushed to avoid garbled lines
• updated guide to the Cplant runtime environment
• change to MCP to remove debug code

To access the service node for purposes of seeing the directory structures available to you, you can "ssh service-22". All of the available filesystem are visible from this system. A df command will show the filesystem local to service-22, and the other available filesystem via an nfs mount. For now to get files off of alaska, you will have to ssh to service-22 and ftp from service-22 to the system you wish to transfer data to. Ftp to service-22 from other systems should be available soon. Note that all users of alaska already have a directory named their username in all filesystems available through fyod. Please direct questions to alaska-help@sandia.gov