Hi Tom,
> True. My point is that the AIX hardware available to our community
> is so slow as to make working on this patch quite unpleasant.
> If what OSUOSL has is not representative of typical AIX installations,
> why not, and could we have access to a machine that is?
We ran the patches on our local nodes, using Meson (with TAP tests). This node
is equipped with two dedicated processors, each with 8 SMT threads, and
dedicated
disks. All steps completed in approximately 12 minutes. Below are the results.
We are currently working with the OSU lab coordinators to secure a node
with similar specifications for the PostgreSQL community development
and testing activities. We will update you.
## time meson setup build -Dtap_tests=enabled -Ddefault_library=shared
-Dreadline=disabled &> meson.out.cl02.new
real 0m45.840s
user 0m8.297s
sys 0m5.811s
## meson configure build/ | grep -i tap
tap_tests enabled [enabled,
disabled, auto] Enable TAP tests
## time ninja
[2306/2306] Linking target src/test/modules/worker_spi/worker_spi.so
real 2m47.907s
user 4m56.973s
sys 0m27.577s
## time ninja test
361/362 icu - postgresql:icu/010_database
OK 6.20s 8 subtests passed
362/362 subscription - postgresql:subscription/100_bugs
OK 43.91s 17 subtests passed
Ok: 331
Fail: 0
Skipped: 31
Full log written to /home/pgdev/postgres/build/meson-logs/testlog.txt
real 8m11.818s
user 1m47.928s
sys 2m16.477s
## prtconf
System Model: IBM,9040-MR9
Machine Serial Number: 132944X
Processor Type: PowerPC_POWER9
Processor Implementation Mode: POWER 9
Processor Version: PV_9_Compat
Number Of Processors: 2
Processor Clock Speed: 3300 MHz
CPU Type: 64-bit
Kernel Type: 64-bit
LPAR Info: 17 urus-cl02
Memory Size: 4096 MB
Good Memory Size: 4096 MB
Platform Firmware level: VM950_175
Firmware Version: IBM,FW950.D0 (VM950_175)
Console Login: enable
Auto Restart: true
Full Core: true
NX Crypto Acceleration: Capable and Enabled
In-Core Crypto Acceleration: Capable, but not Enabled
Recently, we noticed that something seems to be broken in the GNU
build—specifically, 'gmake check‑world' is failing to build, whereas the same
issue does not occur with the Meson build. We’d like to understand whether any
fixes were made on your side in your testing. So that
we can retest the GNU build and share the corresponding results.
Thank you for testing and reporting the performance issues on your node.
Warm regards,
Sriram.
