Collision Rate

<region_name>: collision rate (<percentage>%), <number_of> collisions, <number_of> accesses on region <region_ID>

Explanation

The rate of lock collisions is high.

Tasks that run in different threads are attempting to access a global storage area in parallel. The synchronization required for this often leads to an increased collision rate.

One exception to this in liveCache instances is high collision rates in the OMSVDIR and CNSTVIEW regions (<region_id>). This is normal for certain actions, such as a simultaneous CIF queue transfer.

User Response

You need to act if the collision rate rises above 10%. Generally, the risk of collision rises with the number of processors used (MAXCPU general database parameter). In multiprocessor systems, you should therefore check whether the database system can fulfill the needs of the applications with fewer CPUs.

If high collision rates occur in multiprocessor central systems (database system and application running on the same computer), check whether the computer’s CPU is overloaded, and whether the database threads are blocked by other applications. In this case, the database threads that contain user tasks should receive REAL TIME PRIORITY from the operating system. To avoid operating system blocks however, the value of MAXCPU must be at least one lower than the number of actual CPUs.

Additional Actions

·        If the high collision rates occur in the DATAn, SPLITn or TREEn regions, increase the values of both the general database parameter CACHE_SIZE and the special database parameters _DATA_CACHE_RGNS and _TREE_RGNS.

·        If the high collision rates occur in the TRACE or BUFWRTR regions, then activate the database trace temporarily for troubleshooting only.