| Oracle FAQ | Your Portal to the Oracle Knowledge Grid | |
 |
 | |||
| Oracle FAQ | Your Portal to the Oracle Knowledge Grid | |
|
| |||
Home -> Community -> Mailing Lists -> Oracle-L -> Re: [oracle-l] Re: The Holy War: Disks
Cache can be good if you have short spikes and/or lots of repetition (ie
waste). Sustained, intelligent use of data will not benefit from a cache
- in fact, all the extra code path, abstraction layers, issues, etc.
will offset the investment most of the time. But how many caches do we
really need? We easily have four of them in play these days, all trying
to do the same thing to the same data. That's overhead.
It must be time for a junk-rhyme:
Save your cash
Don't buy cache
Mogens
Connor McDonald wrote:
>SAN cache is one of those great things that
>
>- looks really good on the brochure
>- *might* be really good on your system
>- might *not* be useful on your system
>
>After all, you're basically hoping to find data that is not popular enough to be in the Oracle
>buffer cache, but is popular enough to be happily residing in the SAN cache....Hmmmm
>
>I've send a few places where they just ended up using most of their (san) cache for redo speedup.
>
>All of which comes back to the time old rule...benchmark, benchmark, benchmark :-)
>
>hth
>connor
>
> --- Paul Drake <discgolfdba_at_yahoo.com> wrote: > --- "Koivu, Lisa" <Lisa.Koivu_at_Cendant-TRG.com>
>wrote:
>
>
>>>Hello everyone,
>>>
>>>Windows 2003, v9204, 2 Clariion SANs
>>>
>>>
>>I have a site with a CX400 in use, with W2K3 Server,
>>but its not yet in production, so there isn't much IO
>>to look at. I might be able to run a few tests there.
>>
>>Its a small oltp database and its clone.
>>
>>
>>
>>>I have just been given ~1tb of disk on a new SAN.
>>>The engineer wanted
>>>to give me 3 huge (maxed out) disks, 2 350GB and a
>>>third with the
>>>remainder. I argued for 6 disks similarly sized.
>>>
>>>My fellow DBA supported my argument. The engineer
>>>and the dw architect
>>>wanted 3 disks.
>>>
>>>I am going to have i/o problems no matter what.
>>>Concatenating 10
>>>physical disks into 1 logical disk is going to have
>>>as much i/o latency
>>>percentagewise as 6 physical disks concatenated into
>>>1 logical disk.
>>>
>>>Each disk has 1 bus, so to speak. These buses are
>>>concatenated together
>>>into 1 "device". (I'm being told that "device" is a
>>>unix term and it
>>>doesn't apply in Windows).
>>>
>>>
>>A LUN is a LUN is a LUN.
>>It is apportioned within the management software the
>>same way.
>>
>>
>>
>>>So, concatenating 10
>>>disks (and buses)
>>>together for 1 high speed disk is going to result in
>>>having even more
>>>data on the other side of the "straw".
>>>
>>>
>>>
>>Lisa, did you hike the max IO_size for the operating
>>system? It was 256 KB on w2k by default, I did not
>>check it on w2k3 server. Connor McDonald had a
>>referereference to the setting some time ago. I think
>>I have an article related to hiking this in the
>>regsitry, a max size of 1 MB was possible.
>>
>>Is the oracle server multipathed over multiple
>>Fibrechannel host bus adapters, or at least over
>>multiple ports? If not, its likely that even a 2 Gbps
>>connection will be the rate limiting factor, if you
>>have 30 drives mounted on 3 different buses. You
>>probably wanted multipathing for availability purposes
>>anyways.
>>
>>
>>
>>>I strongly feel if I have 3 disks instead of 6, my
>>>options for
>>>alleviating i/o contention are very limited.
>>>
>>>
>>If your max IO size is 256 KB, then having more mount
>>points should increase throughput. Datawarehouse -
>>you'll likely want 1 MB reads
>>(db_block_size * db_file_multiblock_read_count = 1
>>MB).
>>
>>
>>
>>
>>>Any i/o balancing would be
>>>messier and more difficult. We are going with 6
>>>disks instead of 3 with
>>>the understanding that when we add more disk to this
>>>server, we'll
>>>evaluate performance of the 6 disks and reconsider.
>>>
>>>
>>what is the degree of parallelization that will be
>>used? I'm assumming that you'll only have a 4-way box,
>>that may appear as an 8-way box with SMT enabled.
>>
>>
>>
>>>As far as I know, the Clariion SANs don't have the
>>>whizbang
>>>functionality of the Symmetrix that allows moving
>>>datafiles at the
>>>physical level within the SAN to alleviate i/o
>>>hotspots. I also don't
>>>buy the argument that the SAN cache should alleviate
>>>i/o problems.
>>>
>>>
>>As far as large table scans (and other multiblock IO),
>>its more a matter of IF read-ahead is effective.
>>
>>
>>
>>
>>>This
>>>is a data warehouse that has the potential to become
>>>enormous and it
>>>will blow the size of any SAN cache during data
>>>loads, guaranteed.
>>>
>>>
>>good article in SysAdmin magazine several months ago,
>>comparing cache-centric vs. throughput-centric
>>external storage units. This unit should be good in
>>the non-cache-centric arena.
>>
>>
>>
>>>Is anyone in this type of environment? What have
>>>your experiences been?
>>>Any and all comments are welcome.
>>>
>>>
>>>Thank you
>>>
>>>Lisa Koivu
>>>Senior Monkey
>>>
>>>Cendant Timeshare Resort Group
>>>Orlando, FL, USA
>>>
>>>
>>Pd
>>
>>
>>__________________________________
>>Do you Yahoo!?
>>Yahoo! SiteBuilder - Free web site building tool. Try it!
>>http://webhosting.yahoo.com/ps/sb/
>>-------------------------------------------------------------
>>Please see the official ORACLE-L FAQ: http://www.orafaq.com
>>-------------------------------------------------------------
>>To unsubscribe send email to: oracle-l-request_at_freelists.org
>>put 'unsubscribe' in the subject line.
>>-------------------------------------------------------------
>>
>>
>
>=====
>Connor McDonald
>Co-author: "Mastering Oracle PL/SQL - Practical Solutions" - available now
>web: http://www.oracledba.co.uk
>web: http://www.oaktable.net
>email: connor_mcdonald_at_yahoo.com
>
>"GIVE a man a fish and he will eat for a day. But TEACH him how to fish, and...he will sit in a boat and drink beer all day"
>
>________________________________________________________________________
>BT Yahoo! Broadband - Free modem offer, sign up online today and save £80 http://btyahoo.yahoo.co.uk
>-------------------------------------------------------------
>Please see the official ORACLE-L FAQ: http://www.orafaq.com
>-------------------------------------------------------------
>To unsubscribe send email to: oracle-l-request_at_freelists.org
>put 'unsubscribe' in the subject line.
>-------------------------------------------------------------
>
>
>
-- Archives are at http://www.freelists.org/archives/oracle-l/ FAQ is at http://www.freelists.org/help/fom-serve/cache/1.html -----------------------------------------------------------------Received on Sat Feb 07 2004 - 05:04:44 CST
 |
 |