Saturday, September 27, 2008

NUMA after 10.2.0.4 Upgrade is Automatically Enabled

On HP UX platform I recently did an upgrade project of a 1TB warehouse system from 10.2.0.2 to 10.2.0.4. After upgrade I saw a strange behaviour that Oracle Database alert.log started reporting ORA-07445:[$PXNMOVE] and ORA-07445:[$COLD_KKERFP()+1984]. These errors are diagnosed as faulty CPU , so had to replace all Cell Boards on the Faulty HP Superdome Hardware.

Later the performance went down and we could not load the data on time. Things went bad.
All I saw is the first three CPUs were used by the system and not the rest of the CPUs.




With a lot of investigation between HP & Oracle , it was finally discovered to be the NUMA setup within Oracle Code to use the first cell board and ignore the rest.

alert.log looked like the below for NUMA configuration
and NUMA PG = 3, CPUs = 3


Statement from Oracle was "NUMA is turned on by default 10.2.0.4 and as per the observation only 3 CPU is used."

Later the Init.ora parameter were setup to disable the NUMA.
_enable_numa_optimization=FALSE
_db_block_numa=1
Database was restarted and we were able to see the Oracle processes working on all the 10 CPUS.

Now the alert.log will showed like
and NUMA PG = 1, CPUs = 10

I hope this information will help HP Customers running Oracle Database.

Popular Posts