Message-ID: <4445A1DF.6000705@elde.net>
Date: Wed, 19 Apr 2006 04:35:11 +0200
From: Terje Elde <terje@elde.net>
User-Agent: Thunderbird 1.5 (Windows/20051201)
MIME-Version: 1.0
To: "Jim C. Nasby" <jnasby@pervasive.com>
Cc: pgsql-performance@postgresql.org
Subject: Re: Blocks read for index scans
References: <85979594-1CCA-43C5-A8F2-08685CF41ACE@pervasive.com>
	<443F3BB3.30404@elde.net> <20060418200641.GG49405@pervasive.com>
In-Reply-To: <20060418200641.GG49405@pervasive.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Jim C. Nasby wrote:
>> That said, it's the transactions against disk that typically matter.  On 
>> FreeBSD, you can get an impression of this using 'systat -vmstat', and 
>> watch the KB/t column for your drives.
>>     
>
> On a related note, you know of any way to determine the breakdown
> between read activity and write activity on FreeBSD? vmstat, systat,
> iostat all only return aggregate info. :(
>   


Can't think of a right way to do this ATM, but for a lab-type setup to 
get an idea, you could set up a gmirror volume, then choose a balancing 
algorithm to only read from one of the disks.  The effect should be that 
writes go to both, while reads only go to one.  Activity on the 
write-only disk would give you an idea of the write activity, and 
(read/write disk - write-only disk) would give you an idea of the 
reads.  I have to admit though, seems like quite a bit of hassle, and 
I'm not sure how good the numbers would be, given that at least some of 
the info (KB/transaction) are totals, it'd require a bit of math to get 
decent numbers.  But at least it's something.

Terje