On Fri, Jan 30, 2015 at 6:38 PM, Jim Nasby <Jim.Nasby@bluetreble.com> wrote:
> On 1/30/15 11:08 AM, Robert Haas wrote:
>> The final patch attached her (parallel-dummy-v2.patch) has been
>> updated slightly to incorporate some prefetching logic. It's still
>> just demo code and is not intended for commit. I'm not sure whether
>> the prefetching logic can actually be made to improve performance,
>> either; if anyone feels like playing with that and reporting results
>> back, that would be swell.
>
> Wouldn't we want the prefetching to happen after ReadBuffer() instead of
> before? This way we risk pushing the buffer we actually want out, plus if
> it's not already available the OS probably won't try and read it until it's
> read all the prefetch blocks.
Please feel free to try it both ways and let me know what you find
out. We've had a decent amount of commentary on this patch and the
parallel sequential scan patch, but AFAICT, not much actual testing.
The best way to find out whether your proposal would be an improvement
is to try it. I have my own theory based on the testing I've done so
far, but what would be better than a theory is some evidence,
preferably collected by diverse users on diverse systems.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company