Re: share a new 11gR2 feature

From: Nuno Souto <dbvision_at_iinet.net.au>
Date: Sun, 06 Sep 2009 13:51:44 +1000
Message-ID: <4AA331D0.1080605_at_iinet.net.au>



Matthew Zito wrote,on my timestamp of 6/09/2009 3:24 AM:
> Nuno,
> 
> I think we're conflating the two different points.  I was 

> commenting only on the ability to tell ASM to put certain
> objects closer to the outside of the disks.
>
> In almost all of today's arrays, there is a non-relational
> association between the physical disk and the logical devices
> that are presented to the end host. In addition, having large
> amounts of cache and write re-ordering mitigates the performance
> benefits of placing data on the outside of the disk.
>
> I was just making the point that this is largely inapplicable to
> people who have traditional arrays, and seems to be a feature
> designed to help encourage people that they don't need
> traditional storage arrays.

>
> I agree that removing the requirement to allocate data for empty
> objects is a good thing.

Sorry, I thought you were commenting on the deferred segment allocation. Hence why it didn't make any sense to me.

Yes, I hear you. I'm fully aware of the impact of write-reordering, cache and such on a SAN disk array. In fact, it is one of the great advantages of the SAN: not having to worry about all that.

My comment was more directed at the EMC support personnel that shows up at our place: they constantly ask why aren't we "spreading the database files" across multiple disk stripes, and claiming that we "need to maximize IOPS". It doesn't even cross their minds to actually EXAMINE the characteristics and setup of out SAN and database before they start talking, all they do is disgorge the manuals upfront.

Had they bothered to even LOOK at our setup, they'd realize we have spread the I/O over 24 disks in 3 RAID10 stripes, we don't give a fig about IOPS as our I/O is mostly sequential high volume: bandwidth is the criteria we shoot for. As EMC's own doco states: tune for IOPS with OLTP, tune for bandwidth with DW and similar workloads.

Guess what we run? A DW. Even though we followed all the recommendations from EMC and have no issues whatsoever with IO capacity, every single time an EMC "expert" visits, we have to suffer the interminable sessions of "have you done this-have you done that". It'd give EMC a much better profile in our organization if their "experts" actually LOOKED first before talking, I can assure you.

But sorry, this is just me venting and has nothing to do with what you said. I do agree that to a large extent, those features of enterprise storage make hardware details not as critical for performance.
That is of course assuming one's SAN administrator doesn't go around rabidly disabling the write cache for all bronze-class disk arrays, but there I go venting again... ;)

-- 
Cheers
Nuno Souto
in sunny Sydney, Australia
dbvision_at_iinet.net.au
--
http://www.freelists.org/webpage/oracle-l
Received on Sat Sep 05 2009 - 22:51:44 CDT

Original text of this message