"Jane" <ja**********@i2.com> wrote in message
news:75**************************@posting.google.c om...
Our db2diag.log is full of messages like this:
2004-05-31-17.15.10.383766 Instance:tminst1 Node:000
PID:394948(db2agent (TMDB1) 0) TID:1 Appid:GA140956.EF26.03A4B1202647
data management sqldEscalateLocks Probe:3 Database:TMDB1
ADM5502W The escalation of "4759" locks on table "I2TM .SHPM_T" to
lock intent "S" was successful.
The message is always about Shared locks. What can I do to eliminate these
errors?
Do we need to changes these parameters?
Interval for checking deadlock (ms) (DLCHKTIME) = 10000
Percent. of lock lists per application (MAXLOCKS) = 10
Lock timeout (sec) (LOCKTIMEOUT) = -1
Thanks,
-Jane
That is a warning message, not an error message. It indicates that row
locking has been escalated to table locking. Multiple S (share) table locks
can exist without any contention among different applications, unless
another application needs to update the table at the same time.
You can discourage lock escalation (from row to table) by increasing the
LOCKLIST memory size (DB2 will start escalation when the LOCKLIST is full
with other locks) and increasing MAXLOCKS (the percent of the LOCKLIST that
any one application can use before escalation occurs). If you increase
LOCKLIST memory, you should also increase DBHEAP by the same amount
(LOCKLIST memory comes out of DBHEAP).
Lock escalation to table level can improve performance (because DB2 does not
need to spend time locking and unlocking each row), but can create
contention with other applications if they need to do updates. Lock
escalation to table level is usually a benefit in an ad-hoc query
environment (so long as loads are done when others are not querying). In an
OLTP environment, lock escalation can be a real problem.
LOCKTIMEOUT -1 means that no application waiting for a lock will ever time
out with a -911 reason code 68, it will wait indefinitely. Most people set
this to somewhere between 10 and 60 seconds.
DLCHKTIME indicates how often DB2 checks for deadlocks, which is a slightly
different occurrence than a locktimeout. That looks OK.