Re: Parallel Seq Scan vs kernel read ahead

On Wed, Jun 10, 2020 at 6:04 PM David Rowley <dgrowleyml(at)gmail(dot)com> wrote:
>
> On Wed, 10 Jun 2020 at 17:39, David Rowley <dgrowleyml(at)gmail(dot)com> wrote:
> >
> > On Wed, 10 Jun 2020 at 17:21, Thomas Munro <thomas(dot)munro(at)gmail(dot)com> wrote:
> > > I also heard from Andres that he likes this patch with his AIO
> > > prototype, because of the way request merging works. So it seems like
> > > there are several reasons to want it.
> > >
> > > But ... where should we get the maximum step size from? A GUC?
> >
> > I guess we'd need to determine if other step sizes were better under
> > any conditions. I guess one condition would be if there was a LIMIT
> > clause. I could check if setting it to 1024 makes any difference, but
> > I'm thinking it won't since I got fairly consistent results on all
> > worker settings with the patched version.
>
> I did another round of testing on the same machine trying some step
> sizes larger than 64 blocks. I can confirm that it does improve the
> situation further going bigger than 64.
>

Can we try the same test with 4, 8, 16 workers as well? I don't
foresee any problem with a higher number of workers but it might be
better to once check that if it is not too much additional work.

--
With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com