Class 3 suspensions

Class 3 suspensions indicate potential problems with lock acquisitions, long wait times for synchronous I/O, and long service task switch times.

If the class 3 times are not high, the application wait times could be caused by CPU queuing or by OS/390® paging. This should be shown as class 2 not accounted time.

When evaluating class 3 suspensions, check the following information.
Commit interval
This is calculated as:
COMMIT INTERVAL=CLASS 2 ELAPSED TIME /(COMMITS+ROLLBACKS)

A value of 1 to 5 is acceptable. For example, if CLASS 2 ELAPSED TIME is 1 second and COMMIT is 1, the result is 1. Higher values indicate problems with lock acquisition, especially in batch applications.

Class 3 synchronous I/O time per I/O event
Typically, this time should be about 20 to 30 milliseconds. A longer time indicates a problem because applications are waiting for synchronous I/O.

You can reduce I/O contention by increasing the size of your buffer pools or by reorganizing table and index spaces using REORG.

I/O problems are often system related. If a corresponding Statistics report also indicates problems, take the appropriate action. There might also be problems which could be identified by RMF. Verify the buffer pool data set Statistics block of the report.

Service task switch
The accumulated wait time from switching synchronous execution units, by which Db2 switches from one execution unit to another.
The most common contributors to service task suspensions are:
  • Wait for commit processing for updates (UPDATE COMMIT)
  • Wait for OPEN/CLOSE service task (including HSM recall)
  • Wait for SYSLGRNG recording service task
  • Wait for data set extend/delete/define service task (EXT/DEL/DEF)
  • Wait for other service tasks (OTHER SERVICE)
The significant fields here are the times used by:
  • UPDATE COMMIT

    This is the average service time associated with a Db2 commit, abort, or deallocation. Normally this value should be below 12 milliseconds. If this value is significantly higher, verify the application or raise the interval between commits.

  • OPEN/CLOSE

    If the average time for an OPEN/CLOSE exceeds 150 milliseconds, DSMAX can be too low or too many data sets are open in the system. Adjust the DSMAX threshold. A figure of between 8000 and 10000 is a good working size in most cases.

    Ensure that applications use the correct QUERYTYPE. Set the CLOSERULE parameter of the CREATE TABLESPACE, CREATE INDEX, ALTER TABLESPACE, and ALTER INDEX commands to YES for table spaces and indexes used mainly by batch and NO for those used mainly online.