Обсуждение: pg_restore performance on solaris 10/6

Поиск
Список
Период
Сортировка

pg_restore performance on solaris 10/6

От
Thomas Burns
Дата:
Hi,

We recently started using a Sun x4200 with Solaris 10/6 (the release
from jun of 2006) and postgresql 8.1.3.  I think I may need to do some
tuning, but have no idea what to change.  Using pg_restore to restore
our database literally takes 8 times longer on the x4200 as it does on
OS X (the x4200 hardware should be considerably faster -- it has better
drives, more RAM, faster processors and 4 cores vs 2 cores).  The
loadavg
on the x4200 during the restore is extremely low -- about 0.01 (on OS X,
it would be more like 1.5 or 2.0).  Any idea what could be going on?

Thanks,
Tom Burns



Re: pg_restore performance on solaris 10/6

От
Vivek Khera
Дата:
On Jul 31, 2006, at 3:45 PM, Thomas Burns wrote:

> our database literally takes 8 times longer on the x4200 as it does on
> OS X (the x4200 hardware should be considerably faster -- it has
> better

for a restore like this, bump up the value of checkpoint_segments to
some large value (I use 256 -- but I have a dedicated partition for
the pg_xlog directory which gets big).

also, the normal tuning of the shared memory settings apply.  how
much RAM do you have and what kind of disks are you using?



Вложения

Re: pg_restore performance on solaris 10/6

От
Vivek Khera
Дата:
On Aug 2, 2006, at 8:05 PM, Thomas Burns wrote:

> I have 50000 shared buffers and 10000 temp buffers (the machine has
> 16G ram).   The disks are new/fast SCSI drives

what's their configuration? RAID? if so, what controller and setup?
if not, what controller and setup?

without all the details of your setup, it is impossible to offer
suggestions.


Вложения

Re: pg_restore performance on solaris 10/6

От
Thomas Burns
Дата:
On Aug 1, 2006, at 8:52 AM, Vivek Khera wrote:

>
> On Jul 31, 2006, at 3:45 PM, Thomas Burns wrote:
>
>> our database literally takes 8 times longer on the x4200 as it
>> does on
>> OS X (the x4200 hardware should be considerably faster -- it has
>> better
>
> for a restore like this, bump up the value of checkpoint_segments
> to some large value (I use 256 -- but I have a dedicated partition
> for the pg_xlog directory which gets big).

will give that a shot

>
> also, the normal tuning of the shared memory settings apply.  how
> much RAM do you have and what kind of disks are you using?

I have 50000 shared buffers and 10000 temp buffers (the machine has
16G ram).   The disks are new/fast SCSI drives

>
>

Thomas E. Burns
Co-Founder/CTO, Schoolloop.com
http://www.schoolloop.com
tombu@schoolloop.com
415.255.7285