DomainKey-Signature: a=rsa-sha1; c=nofws;
        d=gmail.com; s=beta;
        h=received:message-id:date:from:sender:to:subject:cc:in-reply-to:mime-version:content-type:content-transfer-encoding:content-disposition:references:x-google-sender-auth;
        b=ga0jxuNijgcdCizCHOQOD3eKOAB9PAN9xvnmtp3Q3i/tt7ZxohmSLAjBxvh4XS3/Wt75WnnldvhbWEO5kf66GywKyljiSxgCewLigmpP2VwIOV4o9xDNY9G9vxXl7N37UMXKb89cC5Rx3M8rJB1xW6jBlSIAaCkSp4d84XQdlFs=
Message-ID: <c2213f680709181942l79088c71x7d4140109a584509@mail.gmail.com>
Date: Tue, 18 Sep 2007 22:42:37 -0400
From: "Alex Gorbachev" <ag@oracloid.com>
To: "Don Seiler" <don@seiler.us>
Subject: Re: Weird database hanging
Cc: oracle-l@freelists.org
In-Reply-To: <716f7a630709181902t73b910a1re2b516e9186101ca@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1
Content-Disposition: inline
References: <20070915071419.33C63743CB5@turing.freelists.org>
	 <009601c7f773$0d30f340$0200a8c0@Primary>
	 <716f7a630709170850w66347a8fr6f1099fbb1bda3e5@mail.gmail.com>
	 <716f7a630709170938h3403bb0dv32110c5893cf998@mail.gmail.com>
	 <716f7a630709181520p7ee8f2f4m9d4bfa7207c65d47@mail.gmail.com>
	 <c2213f680709181650y4b982fe7n2d3375a2379b0efb@mail.gmail.com>
	 <716f7a630709181902t73b910a1re2b516e9186101ca@mail.gmail.com>
Sender: oracle-l-bounce@freelists.org
Errors-to: oracle-l-bounce@freelists.org
Precedence: normal
Reply-to: ag@oracloid.com

Could you paste content of /proc/meminfo?

I'm not sure not using huge pages can cause such behavior but 16 GB
SGA without huge pages on Linux is a killer. Normally, memory is
managed in small 4k blocks. Each process has it's own page table where
for each page it has location, permissions, and etc. For shared
memory, each process will have its own entry in page table but only
when it first accesses it. What can happen with large SGA, default
page size and many Oracle processes - page table can actually occupy
more space than SGA in addition to overhead of on-the-fly allocation
of entries in page table.

That being said, I'm not sure if that's relevant to your case.
If you get qualified support engineer than s/he could hopefully see a
clue there.

Another option could be to OS trace the processes with high activity
during the hang but I'd be careful on production. On the other hand -
it's not working anyway at those times. :)


On 9/18/07, Don Seiler <don@seiler.us> wrote:
> 5. We are not using huge pages.  My SA says he'll set it up, since it
> can't hurt now.  I'm reading the Puschitz tuning page about it now.  I
> suppose shrinking the SGA or shared pool is always an option.
>
> One other difference between old and new servers is that I have
> enabled the default degree of parallelism on our "warehouse" tables
> and indexes, and set NOPARALLEL on our frontend objects.  On our old
> hardware it was kind of random which tables or indexes had it enabled
> and what degree was set.  Also our parallel_max_servers is 32, where
> before it was 8.  I don't think it would hurt to bring it down again.


-- 
Alex Gorbachev, Oracle DBA Brewer, The Pythian Group
http://www.pythian.com/blogs/author/alex http://www.oracloid.com
BAAG party - www.BattleAgainstAnyGuess.com
--
http://www.freelists.org/webpage/oracle-l