Обсуждение: [ADMIN] Help bad results with pgbench

On Mon, Apr 10, 2017 at 9:09 AM, Lazaro Garcia <lazaro3487@gmail.com> wrote:
> Good morning everyone.
>
> I'm having the following problem with pgbench and test again:
>
> I create a database with a scale of 200 and execute a test (read only) with
> 40 users and this is the result:
>
> pgbench -U postgres -p 5433 -j 10 -c 40 -T 30 -h 127.0.0.1 -S  pgbench
> tps = 89329.960851 (including connections establishing)
> tps = 89348.642359 (excluding connections establishing)
>
> Then restart the server and when I run the test again the result is really
> very bad and the IO starts to climb (here the weird thing is that I'm not
> writing just reading)
>
> tps = 229.842173 (including connections establishing)
> tps = 229.870409 (excluding connections establishing)
>
> Can someone give me some suggestions so I can determine what's going on?
>
> PostgreSQL 9.6.2.
>
> Thank you very much for your time.

Yeah this is the file system cache warm vs cold. In the first instance
everything fits in RAM (or most of it) and is there because it's been
accessed. When you reboot the machine is "warming up" so to speak the
file system cache. Often just running select * from table is all you
need to warm them up yourself.

--
To understand recursion, one must first understand recursion.

On Mon, Apr 10, 2017 at 9:48 AM, Scott Marlowe <scott.marlowe@gmail.com> wrote:
> On Mon, Apr 10, 2017 at 9:09 AM, Lazaro Garcia <lazaro3487@gmail.com> wrote:
>> Good morning everyone.
>>
>> I'm having the following problem with pgbench and test again:
>
> Yeah this is the file system cache warm vs cold. In the first instance
> everything fits in RAM (or most of it) and is there because it's been
> accessed. When you reboot the machine is "warming up" so to speak the
> file system cache. Often just running select * from table is all you
> need to warm them up yourself.

So -i 200 is only a couple of gigs. You should fill up the file system
cache pretty fast.

Just keep running pgbench and the numbers should climb over time.

Or you can look into the pg_prewarm module.

https://www.postgresql.org/docs/9.4/static/pgprewarm.html

--
To understand recursion, one must first understand recursion.

Thank you very much for the answer, your righ testing it again durin 30 min and the tps was increasing over the time.

Regards and thank you for the help.


-----Mensaje original-----
De: Scott Marlowe [mailto:scott.marlowe@gmail.com]
Enviado el: lunes, 10 de abril de 2017 11:56 a. m.
Para: Lazaro Garcia
CC: pgsql-admin@postgresql.org
Asunto: Re: [ADMIN] Help bad results with pgbench

On Mon, Apr 10, 2017 at 9:48 AM, Scott Marlowe <scott.marlowe@gmail.com> wrote:
> On Mon, Apr 10, 2017 at 9:09 AM, Lazaro Garcia <lazaro3487@gmail.com> wrote:
>> Good morning everyone.
>>
>> I'm having the following problem with pgbench and test again:
>
> Yeah this is the file system cache warm vs cold. In the first instance
> everything fits in RAM (or most of it) and is there because it's been
> accessed. When you reboot the machine is "warming up" so to speak the
> file system cache. Often just running select * from table is all you
> need to warm them up yourself.

So -i 200 is only a couple of gigs. You should fill up the file system cache pretty fast.

Just keep running pgbench and the numbers should climb over time.

Or you can look into the pg_prewarm module.

https://www.postgresql.org/docs/9.4/static/pgprewarm.html

--
To understand recursion, one must first understand recursion.