To: "Joshua D. Drake" <jd@commandprompt.com>
cc: Xiaoning Ding <dingxn@cse.ohio-state.edu>,
	Scott Marlowe <smarlowe@g2switchworks.com>,
	pgsql-performance@postgresql.org
Subject: Re: scalablility problem 
In-reply-to: <460E82ED.8070106@commandprompt.com> 
References: <460D7220.9010302@cse.ohio-state.edu>
	<1175290903.25022.110.camel@state.g2switchworks.com>
	<460E8198.1010604@cse.ohio-state.edu>
	<460E82ED.8070106@commandprompt.com>
Comments: In-reply-to "Joshua D. Drake" <jd@commandprompt.com>
	message dated "Sat, 31 Mar 2007 08:49:01 -0700"
Date: Sat, 31 Mar 2007 12:31:48 -0400
Message-ID: <6434.1175358708@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

"Joshua D. Drake" <jd@commandprompt.com> writes:
>> I use RHEL 4. I can not understand how the scalability related with 
>> shared memory?

> It isn't RHEL4 and shared memory. It is PostgreSQL and shared memory. 
> Things have changed with PostgreSQL since 7.3 (7.3 is really god awful 
> old) that allow it to more effectively access shared memory and thus 
> provide better performance.

Some specifics:

* bufmgr algorithms redesigned to allow larger number of shared buffers
to be used effectively

* bufmgr redesigned to not have a single lock for management of all
shared buffers; likewise for lockmgr

* lots of marginal tweaks such as paying attention to cache line
alignment of "hot" shared data structures

I'm probably forgetting some things but I think the bufmgr and lockmgr
changes were the biggest improvements in this area.

			regards, tom lane