Hi,
I have two physical database servers:
test server (5 years old):
DELL PowerEdge R730xd, 1x Intel(R) Xeon(R) CPU E5-2637 v4 @ 3.50GHz (4C/8T), 192 GB RAM (12x 16GB PC4-17000 - 36ASF2G72PZ-2G1A2) - one NUMA node
production server (half year old):
DELL PowerEdge R740xd, 2x Intel(R) Xeon(R) Gold 6128 CPU @ 3.40GHz (6C/12T), 512 RAM (16x 32GB PC4-21300 - 36ASF4G72PZ-2G6E1) - two NUMA nodes (256 GB RAM on each NUMA node)
I noticed that my test server is faster in some queries (Oracle 19c EE). I made some tests on the same instance (same amount of SGA and PGA) on both servers.
One of simple tests - I executed 500x times simple query - select count(1) from ruiana.RN_BONIT_DILY_PARCEL;
test server - 13.3 seconds
production server - 16.6 seconds - 25% slover than test server
On both instances buffer cache contains:
OBJECT_OWNER OBJECT_NAME OBJECT_TYPE BLOCKS_IN_CACHE MB_IN_CACHE PERCENT_IN_CACHE NUM_OBJECT_BLOCKS MB_OBJECT BLOCK_SIZE
RUIANA RN_BONIT_DILY_PARCEL TABLE 13979 109,2109375 97,51 14336 112 8192192
I am really disappointed, because my 5 years old test server is faster than the new one.
I think that my problem si related to NUMA, because test server is one NUMA node system and production hase two NUMA nodes. If I tried to set processor affinity to oracle.exe, run times really differs (cca 16s to 21s).
I found some documents on MOS, but only for Release 10.1 to 11.1 - Oracle NUMA Usage Recommendation (Doc ID 759565.1), How NUMA Allocates Memory? (Doc ID 780466.1) etc.
It seems to me, that in 19c NUMA support is disabled and Oracle uses some kind of interleaving memory allocation accross all NUMA nodes.
I would like to know if there are some new documents and recommendations regarding NUMA systems (moder servers has often more than one socket). I would like to use the hardware as much as possible.
I here someone who has some experience with NUMA and Oracle?
Best regards
Matěj.