On 02/12/2011 05:33 PM, Noah Misch wrote:
> On Sat, Feb 12, 2011 at 03:42:17PM -0600, Kevin Grittner wrote:
>> In two hours of testing with a 90GB production database, the copy
>> patch on top of HEAD ran 0.6% faster than HEAD for pg_dumpall
>> (generating identical output files), but feeding that in to and
>> empty cluster with psql ran 8.4% faster with the patch than without!
>> I'm going to repeat that latter with more attention to whether
>> everything made it in OK. (That's not as trivial to check as the
>> dump phase.)
>>
>> Do you see any reason that COPY FROM should be significantly
>> *faster* with the patch?
> No. Up to, say, 0.5% wouldn't be too surprising, but 8.4% is surprising. What
> is the uncertainty of that figure?
>
>
We have seen in the past that changes that might be expected to slow
things down slightly can have the opposite effect. For example, see
<http://people.planetpostgresql.org/andrew/index.php?/archives/37-Puzzling-results.html>
where Tom commented:
Yeah, IME it's not unusual for microbenchmark results to move a percent or three in response to any code change at
all,even unrelated ones. I suppose it's from effects like critical loops breaking across cache lines differently
thanbefore.
cheers
andrew