Re: Database Design on RAID-5 based storage like HDS-99xx

MAK wrote:

>I would like to solicit input from the folks who have experience in
>above area.
>
>We are in process of designing database files lay out on RAID-5 based
>storage (Hitachi Data systems - HDS-99XX). The application that's
>going to use the database (of size 400-500GB) has SLO of 99.5% uptime
>(i.e. just 0.5% of unplanned outage) and very high performance
>requirement (< 3 seconds for most screens/Queries). I/O profile of the
>application is 75% Read, 25% write. Most reads are Random Reads (Index
>Reads).
>
>Brief Desc. about HDS 99XX storage and how it's used in our
>environment (And possibly all other shops using HDS)
>----------------------------------------------------------------------------------------------------------------------------------------------------------------------
>
>Each cabinet of HDS-99xx has nos of 4 disk RAID-5 arrays (Called
>Parity Groups). Each parity group is further sliced in to 15 subparts
>called (LDEVs or LUNs) and presented to HOST's LVM (Veritas or AIx'
>LVM ). These Ldevs are concatenated with VM based stripe and formed
>volumes and these volumes are used to create File systems.
>
>I would like to know the level of separation to be used for following
>Oracle Files categories for the maximum availability with optimal
>performance?
>
>1) Data files
>2) Each member of Online Redo log (We are using 2 member per group)
>3) Backup & Archive logs (Oracle's hot backup, we're keep 2 daily full
>copies and its archive log on the disk to reduce the MTTR)
>
>Questions:
>
>1) What's risk of having Item 1 & Item 2 on the same Parity group
>and/or logically different (not in same LVM volume set)? Would you
>recommend this config given the proclaimed (& probably proven by now)
>reliability of storage system like HDS 99xx.
>
>2) What's the risk of having item 1,2,3 all/some combined physically
>( Same Parity Group ) as well as logically ( Same LVM volume set ) ?
>
>3) Has any customer of HDS ever experienced the failure of RAID-5
>group? If can you provide the details.
>
>I know some them would be difficult to quantify but any ballpark
>figures would be helpful. Also we can separate them each category
>physically as well as logically but will waste lots of space and will
>have huge financial impact in this season of tightest budgets ( DO
>MORE WITH LESS !! )
>Thanks in advance for you quick inputs.
>
>Thanks
>
>Mak
>
>

If you are soliciting advice than put this at the top of the pile ... Oracle does not belong on RAID5,
advises customers not to use RAID5 except in one specialized circumstance, and you should read
the documents before you start going down a road that leads to poor performance.

If you stop guessing at what is best and follow the recommendations documented at tahiti.oracle.com
you won't have any problems and you won't be asking these questions.

Addendum: Space is cheap ... waste it. Crash without recovery ... and downed production systems
cost money. Ask you management the cost to the enterprise for every minute of downtime. Then spend
a few hours-worth on a disk array.

-- 
Daniel Morgan
http://www.outreach.washington.edu/ext/certificates/oad/oad_crs.asp
http://www.outreach.washington.edu/ext/certificates/aoa/aoa_crs.asp
damorgan_at_x.washington.edu
(replace 'x' with a 'u' to reply)