Re: POC: Cache data in GetSnapshotData()

Hi,

On 08/29/2017 05:04 AM, Mithun Cy wrote:
> Test Setting:
> =========
> Server configuration:
> ./postgres -c shared_buffers=8GB -N 300 -c min_wal_size=15GB -c
> max_wal_size=20GB -c checkpoint_timeout=900 -c
> maintenance_work_mem=1GB -c checkpoint_completion_target=0.9 -c
> wal_buffers=256MB &
>
> pgbench configuration:
> scale_factor = 300
> ./pgbench -c $threads -j $threads -T $time_for_reading -M prepared -S postgres
>
> The machine has 64 cores with this patch I can see server starts
> improvement after 64 clients. I have tested up to 256 clients. Which
> shows performance improvement nearly max 39% [1]. This is the best
> case for the patch where once computed snapshotData is reused further.
>
> The worst case seems to be small, quick write transactions which
> frequently invalidate the cached SnapshotData before it is reused by
> any next GetSnapshotData call. As of now, I tested simple update
> tests: pgbench -M Prepared -N on the same machine with the above
> server configuration. I do not see much change in TPS numbers.
>
> All TPS are median of 3 runs
> Clients TPS-With Patch 05 TPS-Base %Diff
> 1 752.461117 755.186777 -0.3%
> 64 32171.296537 31202.153576 +3.1%
> 128 41059.660769 40061.929658 +2.49%
>
> I will do some profiling and find out why this case is not costing us
> some performance due to caching overhead.
>

I have done a run with this patch on a 2S/28C/56T/256Gb w/ 2 x RAID10
SSD machine.

Both for -M prepared, and -M prepared -S I'm not seeing any improvements
(1 to 375 clients); e.g. +-1%.

Although the -M prepared -S case should improve, I'm not sure that the
extra overhead in the -M prepared case is worth the added code complexity.

Best regards,
Jesper