Re: ANALYZE sampling is too good

On 10/12/13 13:53, Mark Kirkwood wrote:
> On 10/12/13 13:20, Mark Kirkwood wrote:
>> On 10/12/13 13:14, Mark Kirkwood wrote:
>>> On 10/12/13 12:14, Heikki Linnakangas wrote:
>>>>
>>>>
>>>> I took a stab at using posix_fadvise() in ANALYZE. It turned out to
>>>> be very easy, patch attached. Your mileage may vary, but I'm seeing
>>>> a nice gain from this on my laptop. Taking a 30000 page sample of a
>>>> table with 717717 pages (ie. slightly larger than RAM), ANALYZE
>>>> takes about 6 seconds without the patch, and less than a second
>>>> with the patch, with effective_io_concurrency=10. If anyone with a
>>>> good test data set loaded would like to test this and post some
>>>> numbers, that would be great.
>>>>
>>>>
>>>
>>>
>>>
>>> I did a test run:
>>>
>>> pgbench scale 2000 (pgbench_accounts approx 25GB).
>>> postgres 9.4
>>>
>>> i7 3.5Ghz Cpu
>>> 16GB Ram
>>> 500 GB Velociraptor 10K
>>>
>>> (cold os and pg cache both runs)
>>> Without patch: ANALYZE pgbench_accounts 90s
>>> With patch: ANALYZE pgbench_accounts 91s
>>>
>>> So I'm essentially seeing no difference :-(
>>
>>
>> Arrg - sorry forgot the important bits:
>>
>> Ubuntu 13.10 (kernel 3.11.0-14)
>> filesystem is ext4
>>
>>
>>
>
> Doing the same test as above, but on a 80GB Intel 520 (ext4 filesystem
> mounted with discard):
>
> (cold os and pg cache both runs)
> Without patch: ANALYZE pgbench_accounts 5s
> With patch: ANALYZE pgbench_accounts 5s
>
>
>
>
>

Redoing the filesystem on the 520 as btrfs didn't seem to make any
difference either:

(cold os and pg cache both runs)
Without patch: ANALYZE pgbench_accounts 6.4s
With patch: ANALYZE pgbench_accounts 6.4s