Re: IO performance

From: Stéphane Faroult <sfaroult_at_roughsea.com>
Date: Sat, 24 Oct 2015 07:13:47 -0500
Message-ID: <562B75FB.9090308_at_roughsea.com>



With a big SAN, you usually have a big cache, and many servers connected to the same machine. As long as you find most of what you are looking for in your cache, performance is great. Now, if one of the servers starts a process in which massive scans are performed (let's say massive index rebuilding, or recomputing all stats), it will simply flush the cache, replacing the blocks that other servers would have found there by its own blocks, which means that for the others what Oracle was seeing as a physical I/O but wasn't really one will turn into a true one. Performance degrades on other servers without any apparent reason.   You should see what happens on ALL servers connected to the SAN, and reschedule some operations if they put too much strain on the SAN cache.

HTH, Stéphnae Faroult

On 24/10/15 01:23, Cee Pee wrote:
> Hi
>
> I have a clarification about the IO performance. Let's say we find
> that the IO performance is sub par at certain times in a day from
> performance charts. With the storage all in one big SAN these days,
> how do people really improve the IO performance. I would think that IO
> characteristics of a SAN is 'uniform', so what would be the options to
> improve IO performance.
>
> CP.

--
http://www.freelists.org/webpage/oracle-l
Received on Sat Oct 24 2015 - 14:13:47 CEST

Original text of this message