Re: read stream: Backward I/O combining

From: Nazir Bilal Yavuz <byavuz81(at)gmail(dot)com>
To: Tomas Vondra <tomas(at)vondra(dot)me>
Cc: PostgreSQL Hackers <pgsql-hackers(at)lists(dot)postgresql(dot)org>, Thomas Munro <thomas(dot)munro(at)gmail(dot)com>, Peter Geoghegan <pg(at)bowt(dot)ie>, Andres Freund <andres(at)anarazel(dot)de>
Subject: Re: read stream: Backward I/O combining
Date: 2026-07-24 08:07:26
Message-ID: CAN55FZ0U-q_qvb6-YG=V6Vp6SDX2gY-4D60SuZk3ObJ--JYhEw@mail.gmail.com
Views: Whole Thread | Raw Message | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Hi,

On Mon, 20 Jul 2026 at 17:57, Tomas Vondra <tomas(at)vondra(dot)me> wrote:
>
> Thanks for the patch. I got this patch a couple days/week ago, as part
> of the work on index prefetching. So I got to do a bit of benchmarking.
>
> The pattern is pretty clear. The DESC (backward access) is much slower,
> hence the huge area of red results for runs without index prefetch.
>
> The index prefetch helps quite a bit, with ~5x speedup, which is nice.
> But it's still ~5x slower than the ASC case. The combining gets us to
> ~1.5x compared to master/ASC for all the tested SSD devices etc.
>
> So I think this works pretty well, and it largely addresses the issue
> with backward access.

Thanks for looking into this and thanks for the benchmarks, they are
very helpful!

After an off-list discussion with Andres, we decided it would be
better to share this patch as a version that applies cleanly on top of
master. Although this patch is an index-prefetching specific patch, it
can be tested by the test_aio infrastructure and also can be
benchmarked by the same infrastructure (as done in 0002). So, here is
a v2 which applies cleanly on top of master.

--
Regards,
Nazir Bilal Yavuz
Microsoft

Attachment Content-Type Size
v2-0001-Add-read-stream-backwards-I-O-combining.patch text/x-patch 42.8 KB
v2-0002-Benchmark-script.patch text/x-patch 8.4 KB

In response to

Browse pgsql-hackers by date

  From Date Subject
Next Message Chao Li 2026-07-24 08:38:12 Re: Collect ALTER PUBLICATION commands for event triggers
Previous Message Laurenz Albe 2026-07-24 08:02:45 Re: Build warning with meson and dtrace on Fedora 43