RE: RAID system max throughput

Thanks to everyone that shared their thoughts on this issue.

Steven - Thanks very much for your insight of throughput vs. latency. I think this reduces the concepts to a simple enough level that I can explain them to others. The system administrators also do our wide-area networks and the focus tends to be on throughput, but they have also realized that latency times can also produce poor performance. Dennis Williams
DBA
Lifetouch, Inc.
dwilliams_at_lifetouch.com

-----Original Message-----
Sent: Friday, December 07, 2001 8:40 PM
To: Multiple recipients of list ORACLE-L

• DENNIS WILLIAMS <DWILLIAMS_at_LIFETOUCH.COM>

> Whenever I discuss disk waits with my system administrator, I always get
> the reply that "the RAID system isn't anywhere near its rated
> throughput". Maybe I'm wrong, but I don't see any of the tuning books
> mentioning that as a relevant performance characteristic. However, I've
> never been able to move the discussion beyond this point. Can anyone
> straighten me out on this point or point me to a resource that might be
> applicable.
>
> Our system is Oracle 8.1.6, Compaq Tru64. We use hardware RAID-5 with a
> battery-backed RAM cache, and have about 3 RAID sets (plus some extra
> disks for redo logs, etc.), and performance is fine, but I'm always
> looking as to how we can improve Oracle performance. The application is
> our corporate ERP system.

Two things will zap you on device I/O: bandwidth or latency. Most people look at bandwidth -- same basic numbers for both networking and disks. Latency is basically the turnaround time.

If you screw up the setup of any I/O system then latency can reduce performance to the point where bandwidth is irrelevant. By analogy, you can put concrete tires on a Porsche and go nowhere also.

The real measure of what's going on starts at the O/S level looking at the frequency and duration of proc's in a device wait state (a.k.a. "blocked for I/O") on the disks. If this is minimal then forget it.

You can also end up with screwy results on large shared disk systems due to competition. SAN's can get placed on overloaded network segments; ERP's can easily get hot-spots from various users colliding.

In general RAID5 with a stripe size == system I/O page will perform rather nicely. If your system page sizes vary or the raidset has an offball number of disks (e.g., 6 drives for an 8K page) then you'll take a hit writing extra data to maintain the RAID5 parity.

--
Steven Lembark                               2930 W. Palmer
Workhorse Computing                       Chicago, IL 60647
                                            +1 800 762 1582
-- 
Please see the official ORACLE-L FAQ: http://www.orafaq.com
-- 
Author: Steven Lembark
  INET: lembark_at_wrkhors.com

Fat City Network Services    -- (858) 538-5051  FAX: (858) 538-5051
San Diego, California        -- Public Internet access / Mailing Lists
--------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in
the message BODY, include a line containing: UNSUB ORACLE-L
(or the name of mailing list you want to be removed from).  You may
also send the HELP command for other information (like subscribing).
-- 
Please see the official ORACLE-L FAQ: http://www.orafaq.com
-- 
Author: DENNIS WILLIAMS
  INET: DWILLIAMS_at_LIFETOUCH.COM

Fat City Network Services    -- (858) 538-5051  FAX: (858) 538-5051
San Diego, California        -- Public Internet access / Mailing Lists
--------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in
the message BODY, include a line containing: UNSUB ORACLE-L
(or the name of mailing list you want to be removed from).  You may
also send the HELP command for other information (like subscribing).