Message-ID: <443A2856.6050905@laterooms.com>
Date: Mon, 10 Apr 2006 10:41:42 +0100
From: Gavin Hamill <gdh@laterooms.com>
User-Agent: Mozilla Thunderbird 1.0.7 (X11/20051013)
MIME-Version: 1.0
To: Simon Riggs <simon@2ndquadrant.com>
Cc: pgsql-performance@postgresql.org
Subject: Re: pg 8.1.3, AIX, huge box, painfully slow.
References: <C05C3C44.211A5%llonergan@greenplum.com>	
	<17613.1144451138@sss.pgh.pa.us>
	<1144661209.21409.131.camel@localhost.localdomain>
In-Reply-To: <1144661209.21409.131.camel@localhost.localdomain>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Simon Riggs wrote:

> pSeries cache lines are 128 bytes wide, so I'd go straight to 128.
>
Hello :)

OK, that line of code is:

#define LWLOCK_PADDED_SIZE      (sizeof(LWLock) <= 16 ? 16 : 32)

What should I change this to? I don't understand the syntax of the <= 16 
? : stuff...

would a simple  "#define LWLOCK_PADDED_SIZE  128" be sufficient?

>If you're renting all 8 CPUs, I'd drop to 4 and try that instead. With 8
>CPUs the contention will vary according to what each CPU is doing at any
>one time - when they all hit the contention spot, things will get worse.
>
>  
>
We have a physical machine installed in our rack at the data centre, 
rather than renting a virtual partition of a real machine... I'm not 
sure how to enable/disable CPUs even with the help of 'smitty' :)

>The pSeries has good CPUs and great caching, so I'd expect contention to
>be somewhat more apparent as a bottleneck.
>
>  
>
Yep, I expected 32MB of 'L3' cache would yield impressive results :)

Cheers,
Gavin.