On Mon, 2007-07-23 at 14:19 -0400, Tom Lane wrote:
> "Simon Riggs" <simon@2ndquadrant.com> writes:
> > currPos and markPos are defined as BTScanPosData, which is an array of
> > BTScanPosItems. That makes BTScanOpaqueData up to 8232 bytes, which
> > seems wasteful since markPos is only ever used during merge joins. Most
> > of that space isn't even used during merge joins either, we just do that
> > to slightly optimise the speed of the restore during merge joins.
>
> Ah. I was seeing it as 6600 bytes on HPPA and 6608 on x86_64, but
> I forgot that both of those architectures have MAXALIGN = 8. On a
> MAXALIGN = 4 machine, MaxIndexTuplesPerPage will be significantly
> larger, leading to larger BTScanPosData.
>
> Not sure it's worth fooling with, given that these days almost everyone
> who's seriously concerned about performance is probably using 64bit
> hardware. One less malloc cycle per indexscan is never going to be a
> measurable savings anyway...
Oh sure, I was thinking to avoid Solaris' mutex by avoiding malloc()
completely.
--
Simon Riggs
EnterpriseDB http://www.enterprisedb.com