We are using 10.2.0.3 and not facing any of the BUG mentioned in Oracle NUMA usage recommendation [ID 759565.1]. But our OS & hardware is NUMA enabled. If we disable OS (Linux x86_64) & Hardware (AMD Optron Server on HP Proliant boxes) level then performance decrease significantly.
The best performance we got is with NUMA enabled at every level i.e. Hardware/OS/DB (by
enablenuma_optimization=TRUE). We are not changing any other NUMA init.ora parameter e.g.
dbblock_numa. We have 4 NUMA nodes so do we need to set this
dbblock_numa=4?
numactl --hardware
available: 4 nodes (0-3)
node 0 size: 8056 MB
node 0 free: 651 MB
node 1 size: 8080 MB
node 1 free: 1656 MB
node 2 size: 8080 MB
node 2 free: 1759 MB
node 3 size: 8080 MB
node 3 free: 1694 MB
node distances:
node 0 1 2 3
0: 10 20 20 20
1: 20 10 20 20
2: 20 20 10 20
3: 20 20 20 10
Also even with NUMA enabled at DB level, SGA (24 GB) distribution is not happening evenly on NUMA
nodes. Do we need to upgrade the DB patchset to 10.2.0.4 and apply the Patch 8199533 for effective NUMA usage?
following is the NUMA stat:
numastat
node0 node1 node2 node3
numa_hit 321351567 42297617 61060565 42997951
numa_miss 700232 23301797 37434003 29815244
numa_foreign 86695735 3412301 844487 298753
interleave_hit 7082 25292 28909 29702
local_node 321347157 42236010 60981069 42920729
other_node 704642 23363404 37513499 29892466
Clearly, Node0 is more burdened with memory requests. All the Oracle & other internet forum discussion suggest to upgrade to 10.2.0.4 and apply DB Patch 8199533 for efficient NUMA utilization. We cannot afford to disable NUMA at OS & hardware.