I am stress testing an application that performs massive inserts (logs
tons of data) into Oracle tables using direct path insert. Multiple
tables are used to provide scalability (each log source inserts into a
dedicated table). This application uses direct path inserts and bypasses
the cache so I can see that it also scales in RAC (10g RAC, 6 nodes).
BUT increasing the number of log sources I have now started to see the
following bottleneck: row cache wait (on dc_tablespace_quotas), for
example from a 10046 trace:
---------------------------
WAIT #19: nam='row cache lock' ela= 2888862 cache id=5 mode=0 request=5
obj#=76647 tim=1122164147766228
WAIT #19: nam='row cache lock' ela= 2824967 cache id=5 mode=0 request=5
obj#=76647 tim=1122164150591291
It looks to me as row cache lock contention between direct path inserts,
probably made worse by rac/GES. (Note also that I don't enforce
tablespace quotas (the application owner has unlimited tablespace).
Any ideas on how to tune this?
