Re: Misaligned BufferDescriptors causing major performance problems on AMD

On 2014-12-29 16:59:05 -0500, Bruce Momjian wrote:
> I am glad someone else considers this important.

I do consider it important. I just considered the lwlock scalability
more important...

> Andres reported the above 2x pgbench difference in February, but no
> action was taken as everyone felt there needed to be more performance
> testing, but it never happened:

FWIW, I have no idea what exactly should be tested there. Right now I
think what we should do is to remove the BUFFERALIGN from shmem.c and
instead add explicit alignment code in a couple callers
(BufferDescriptors/Blocks, proc.c stuff).

> 	$ pgbench --initialize --scale 1 pgbench

Scale 1 isn't particularly helpful in benchmarks, not even read only
ones.

> 	$ pgbench --protocol prepared --client 16 --jobs 16 --transactions 100000 --select-only pgbench

I'd suspect you're more likely to see differences with a higher client
count. Also, I seriously doubt 100k xacts is enough to get stable
results - even on my laptop I get 100k+ TPS. I'd suggest using something
like -P 1 -T 100 or so.

Greetings,

Andres Freund

-- Andres Freund	                   http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training & Services