Date: Tue, 24 Oct 2006 14:31:10 +0100
From: Jignesh Shah <J.K.Shah@Sun.COM>
Subject: Re: Copy database performance issue
In-reply-to: <Pine.GSO.4.64.0610231734220.3930@kingcheetah.tanabi.org>
To: Steve <cheetah@tanabi.org>
Cc: pgsql-performance@postgresql.org
Message-id: <453E159E.4000203@sun.com>
MIME-version: 1.0
Content-type: text/plain; format=flowed; charset=ISO-8859-15
Content-transfer-encoding: 7BIT
References: <Pine.GSO.4.64.0610231734220.3930@kingcheetah.tanabi.org>
User-Agent: Thunderbird 1.5.0.7 (Windows/20060909)

Steve,

Are you using the latest update release of Solaris 10 ?

When you are doing the copy, did you check with prstat -amL to see if it 
is saturating on any CPU?

If it is saturating on a CPU then atleast it will narrow down that you 
need to improve the CPU utilization of the copy process.

Brendan Greg's "hotuser" script which uses DTrace and Pearl post 
processing will help you to figure out which functions is causing the 
high CPU utilization and then maybe somebody from the PostgreSQL team 
can figure out what's happening that is causing the slow copy.

If none of the cores show up as near 100% then the next step is to 
figure out if any disk is 100% utilized via iostat -xczmP .

With this information it might help to figure out the next steps in your 
case.

Regards,
Jignesh


Steve wrote:
> Hello there;
>
> I've got an application that has to copy an existing database to a new 
> database on the same machine.
>
> I used to do this with a pg_dump command piped to psql to perform the 
> copy; however the database is 18 gigs large on disk and this takes a 
> LONG time to do.
>
> So I read up, found some things in this list's archives, and learned 
> that I can use createdb --template=old_database_name to do the copy in 
> a much faster way since people are not accessing the database while 
> this copy happens.
>
>
> The problem is, it's still too slow.  My question is, is there any way 
> I can use 'cp' or something similar to copy the data, and THEN after 
> that's done modify the database system files/system tables to 
> recognize the copied database?
>
> For what it's worth, I've got fsync turned off, and I've read every 
> tuning thing out there and my settings there are probably pretty 
> good.  It's a Solaris 10 machine (V440, 2 processor, 4 Ultra320 
> drives, 8 gig ram) and here's some stats:
>
> shared_buffers = 300000
> work_mem = 102400
> maintenance_work_mem = 1024000
>
> bgwriter_lru_maxpages=0
> bgwriter_lru_percent=0
>
> fsync = off
> wal_buffers = 128
> checkpoint_segments = 64
>
>
> Thank you!
>
>
> Steve Conley
>
> ---------------------------(end of broadcast)---------------------------
> TIP 6: explain analyze is your friend