From: | "Blake, Geoff" <blakgeof(at)amazon(dot)com> |
---|---|
To: | Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> |
Cc: | "pgsql-hackers(at)lists(dot)postgresql(dot)org" <pgsql-hackers(at)lists(dot)postgresql(dot)org> |
Subject: | Re: Add spin_delay() implementation for Arm in s_lock.h |
Date: | 2021-12-13 17:27:00 |
Message-ID: | 7E987754-1C43-477B-8FEB-CE894C231B36@amazon.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
Hi Tom,
> What did you test exactly?
Tested 3 benchmark configurations on an m6g.16xlarge (Graviton2, 64 cpus, 256GB RAM)
I set the scale factor to consume about 1/3 of 256GB and the other parameters in the next line.
pgbench setup: -F 90 -s 5622 -c 256
Pgbench select-only w/ patch 662804 tps (-0.5%)
w/o patch 666354 tps.
tcpb-like w/ patch 35844 tps (0%)
w/o patch 35835 tps
We also test with Hammerdb when evaluating patches, it shows the patch gets +3%:
Hammerdb (192 Warehouse 256 clients)
w/ patch 1147463 NOPM (+3%)
w/o patch 1112908 NOPM
I've run pgbench more than once and the measured TPS values overlap, even though the means on select-only show a small degradation at the moment I am concluding it is noise. On Hammerdb, the results show a measurable difference.
Thanks,
Geoff
From | Date | Subject | |
---|---|---|---|
Next Message | Robert Haas | 2021-12-13 17:33:50 | Re: pg_dump versus ancient server versions |
Previous Message | Tom Lane | 2021-12-13 17:23:10 | Re: pg_dump versus ancient server versions |