DB2_SQLWORKSPACE_CACHE may be set in order to improve
performance under highly concurrent OLTP workloads. In an
attempt to avoid an out of memory error condition in either an
application heap or the application shared heap, the internal
workspace cache setting is temporarily reset to 0, overriding
the DB2_SQLWORKSPACE_CACHE setting, and an attempt is made to
remove all cached memory, i.e. workspace that is not currently
in use. This may result in a sudden decrease in performance.
Over time, assuming the memory constraints do not persist, the
cache setting will return to normal (or whatever is specified in
the DB2_SQLWORKSPACE_CACHE setting).
The out of memory condition will more commonly occur in an
environment with a fixed APPLHEAPSZ, APPL_MEMORY, or
INSTANCE_MEMORY setting. Frequently the freeing of cached
memory will not avoid the out of memory condition. Either the
memory allocation is an unusually high single allocation that
will fail regardless, or the memory constraint condition is
pervasive due to an abnormal condition such as a memory leak or
misconfiguration.
The symptoms accompanying the performance degradation may
include an out-of-memory error in the Application Heap or
Application Shared Heap as indicated by the db2diag.log messages
below, as well as a sudden decrease in Application Shared Heap
memory consumption, which can be monitored via
MON_GET_MEMORY_POOL (MEMORY_POOL_TYPE = APPL_SHARED) or db2pd
-db <db> -mempools (PoolName = appshrh).
Out of memory failure for Application Heap (APPLHEAPSZ) on node
0.
Out of memory failure for Applications Shared Heap (application
shared) on node 0. |
