Date: Tue, 20 Apr 2010 13:39:19 -0700
From: David Kerr <dmk@mr-paradox.net>
To: Greg Smith <greg@2ndquadrant.com>
Cc: Scott Marlowe <scott.marlowe@gmail.com>,
	Nikolas Everett <nik9000@gmail.com>, Robert Haas <robertmhaas@gmail.com>,
	pgsql-performance@postgresql.org
Subject: Re: Very high effective_cache_size == worse performance?
Message-ID: <20100420203919.GA62103@mr-paradox.net>
References: <20100420173936.GA50886@mr-paradox.net>
	<h2w603c8f071004201044tbc8ce8fenb846b7643a2b9376@mail.gmail.com>
	<20100420180351.GA51940@mr-paradox.net>
	<w2ld4e11e981004201112v3571f350zaaa0112d1e035097@mail.gmail.com>
	<20100420182030.GA53489@mr-paradox.net>
	<j2qdcc563d11004201123u21219445v4a9129ee7d7b520b@mail.gmail.com>
	<20100420182832.GB53489@mr-paradox.net>
	<4BCE0E0C.1020903@2ndquadrant.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <4BCE0E0C.1020903@2ndquadrant.com>
User-Agent: Mutt/1.4.2.3i

On Tue, Apr 20, 2010 at 04:26:52PM -0400, Greg Smith wrote:
- David Kerr wrote:
- >the db, xlog and logs are all on separate areas of the SAN.
- >separate I/O controllers, etc on the SAN. it's setup well, I wouldn't 
- >expect
- >contention there.
- >  
- 
- Just because you don't expect it doesn't mean it's not there.  
- Particularly something as complicated as a SAN setup, presuming anything 
- without actually benchmarking it is a recipe for fuzzy diagnostics when 
- problems pop up.  If you took anyone's word that your SAN has good 
- performance without confirming it yourself, that's a path that's lead 
- many to trouble.

that's actually what I'm doing, performance testing this environment.
everything's on the table for me at this point. 

- Anyway, as Robert already stated, effective_cache_size only impacts how 
- some very specific types of queries are executed; that's it.  If there's 
- some sort of query behavior involved in your load, maybe that has 
- something to do with your slowdown, but it doesn't explain general slow 
- performance.  Other possibilities include that something else changed 
- when you reloaded the server as part of that, or it's a complete 
- coincidence--perhaps autoanalyze happened to finish at around the same 
- time and it lead to new plans.

Ok that's good to know. I didn't think it would have any impact, and was
surprised when it appeared to.

I just finished running the test on another machine and wasn't able to 
reproduce the problem, so that's good news in some ways. But now i'm back 
to the drawing board.

I don't think it's anything in the Db that's causing it. ( drop and re-create
the db between tests) I actually suspect a hardware issue somewhere. 

Dave