Pipe tuning considerations

The following situations can be identified:

• When participants wait at various synchronization points, the following potential problem areas should be examined or considered:
  • Scheduling problems

  • Job JCL, to verify that the required steps are executing in parallel

    For example, if Step 1 of Job A and Step 5 of Job B are participants of the same pipe, these two steps should run in parallel. If the jobs are submitted concurrently, Step 1 of Job A will be delayed until Step 5 of Job B begins executing.

  • Application logic, to verify that the nature of the applications is similar

    For example, make sure that all participants open the pipe at about the same stage of the process.

  • Deadlocks
• When a writer is waiting too long for I/O, you might need to increase the number of pipe buffers or add an additional reader (if the Reader data option is set to N (next)). If you cannot set the Reader data option to N, consider increasing the I/O wait time limit.
• When a reader is waiting too long for I/O, consider the following points:
  • When the Reader data option is set to A (All), faster readers might wait for I/O because slower readers have not finished reading data from the pipe. If the speed of the readers cannot be balanced, you can create a physical file along with the pipe, remove from the pipe the readers that cause the delay, and let them run later by using the physical file.
  • When the Reader data option is set to A (All), all readers might wait for I/O because the writers are too slow. You can add writers to increase the data transfer rate.
  • When the Reader data option is set to N (Next), some or all of the readers might wait for I/O because the writers are too slow or because there are too many readers. You can reduce the number of readers or add writers to increase the data transfer rate.
  • If the preceding actions cannot be performed, consider increasing the I/O wait time limit.
• If the No-Operation wait time is exceeded, examine the application logic to determine whether large time intervals occur between accesses to the pipe. If this is the case, increase the No-Operation wait time limit.
• Parallel Sysplex considerations include:
  • When Global pipes are used in a Parallel Sysplex environment, the participants run on different systems and use Parallel Sysplex resources, mainly the Coupling Facility (CF). A Global pipe is always slower than a Local pipe (in which all of the participants run on the same system and access pipe buffers directly).
  • When pipe participants (Local or Remote) are running on systems with different CPU capacity, real storage capacity, or CPU utilization, pipe access speed is not balanced. Participants that are running on the stronger or less-used systems might have to wait for the other participants.