On Tue, Nov 15, 2016 at 7:14 AM, Tsunakawa, Takayuki
<tsunakawa.takay@jp.fujitsu.com> wrote:
> From: pgsql-hackers-owner@postgresql.org
>> [mailto:pgsql-hackers-owner@postgresql.org] On Behalf Of Amit Kapila
>> Okay, not a problem. However, I am not sure the results in this thread
>> are sufficient proof as for read-only tests, there is no noticeable win
>> by increasing shared buffers and read-write tests seems to be quite short
>> (60 seconds) to rely on it.
>
> I think the reason why increasing shared_buffers didn't give better performance for read-only tests than you expect
isthat the relation files are cached in the filesystem cache. The purpose of this verification is to know that the
effectiveupper limit is not 512MB (which is too small now), and I think the purpose is achieved. There may be another
threshold,say 32GB or 128GB, over which the performance degrades due to PostgreSQL implementation, but that's another
topicwhich also applies to other OSes.
>
If we don't get any benefit by increasing the shared_buffers on
windows, then what advantage do you see in recommending higher value?
> How about 3 minutes for read-write tests? How long do you typically run?
>
I generally run it for 20 to 30 mins for read-write tests. Also, to
ensure consistent data, please consider changing following parameters
in postgresql.conf
checkpoint_timeout = 35 minutes or so, min_wal_size = 5GB or so,
max_wal_size = 20GB or so and checkpoint_completion_target=0.9.
Apart from above, ensure to run manual checkpoint (checkpoint command)
after each test.
--
With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com