Message-ID: <508163B5.6070902@optionshouse.com>
Date: Fri, 19 Oct 2012 09:29:09 -0500
From: Shaun Thomas <sthomas@optionshouse.com>
Reply-To: <sthomas@optionshouse.com>
Organization: Peak6
User-Agent: Mozilla/5.0 (X11; Linux x86_64;
	rv:16.0) Gecko/20121011 Thunderbird/16.0.1
MIME-Version: 1.0
To: "pgsql-performance@postgresql.org" <pgsql-performance@postgresql.org>
Subject: Tablespace-derived stats?
Content-Type: text/plain; charset="ISO-8859-1"; format=flowed
content-transfer-encoding: 7bit

Hello Perf,

Lately I've been pondering. As systems get more complex, it's not 
uncommon for tiered storage to enter the picture. Say for instance, a 
user has some really fast tables on a NVRAM-based device, and 
slower-access stuff on a RAID, even slower stuff on an EDB, and variants 
like local disk or a RAM drive.

Yet there's only one global setting for random_page_cost, and 
seq_page_cost, and so on.

Would there be any benefit at all to adding these as parameters to the 
tablespaces themselves? I can imagine the planner could override the 
default with the tablespace setting on a per-table basis when 
calculating the cost of retrieving rows from tables/indexes on faster or 
slower storage.

This is especially true since each of the storage engines I listed have 
drastically different performance profiles, but no way to hint to the 
planner. There was a talk at the last PG Open about his EDB tests vastly 
preferring partitioning and sequential access because random access was 
so terrible. But NVRAM has the opposite metric. Currently, tuning for 
one necessarily works against the other.

I didn't see anything in the Todo Wiki, so I figured I'd ask. :)

Thanks!

-- 
Shaun Thomas
OptionsHouse | 141 W. Jackson Blvd. | Suite 500 | Chicago IL, 60604
312-444-8534
sthomas@optionshouse.com

______________________________________________

See http://www.peak6.com/email_disclaimer/ for terms and conditions related to this email