On Sun, Jan 15, 2012 at 4:59 PM, Jeff Janes <jeff.janes@gmail.com> wrote:
>
> The attached patch allows it to reuse that memory. On my meager
> system it reduced the building of an index on an integer column in a
> skinny 200 million row totally randomly ordered table by about 3% from
> a baseline of 25 minutes.
>
Just to give a standard review, this patch is one line change and
applies cleanly, builds ok.
I'm not pretty sure what exactly you're trying to accomplish, but it
seems to me that it's avoiding the first dumptuples cycle by forcing
availMem = 0 even if it's negative. I read your comments as it'd be
avoiding to alternate reading/writing back and force with scattered
memory failing memory cache much during merge phase, but actually it
doesn't affect merge phase but only init-dump phase, does it? If so,
I'm not so convinced your benchmark gave 3 % gain by this change.
Correct me as I'm probably wrong.
Anyway, it's nice to modify the comment just above the change, since
it's now true with it.
Thanks,
--
Hitoshi Harada