Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.tools -> Re: Does RAID 5 contradict and minimize the benefit of OFA on NT?

Re: Does RAID 5 contradict and minimize the benefit of OFA on NT?

From: Jerry Gitomer <jgitomer_at_erols.com>
Date: 2000/04/16
Message-ID: <20000416.3214500@noname.nodomain.nowhere>

>>>>>>>>>>>>>>>>>> Original Message <<<<<<<<<<<<<<<<<<

On 4/15/00, 12:53:33 AM, "George V. Huse, Jr." <buzzhuse_at_flash.net> wrote regarding Re: Does RAID 5 contradict and minimize the benefit of OFA on NT?:

> Howdy, sorry for barging in, but may I ask a few questions?
 

> 1. I take it we should not put the Oracle db files on a RAID5
 volume,
> correct?

        Although not the best configuration RAID5 will work and is frequently used for Oracle databases. The additional I/O activity required in writing data to RAID5 will eventually have a negative impact on overall system performance. If you do go with RAID5 just be sure to inform management that you may have to double the disk configuration in the near future in order to realize acceptable performance.

> 2. The redo log file should be on a separate physical disk from the
 

db &
> also not on a RAID5 volume, correct? I would choose a RAID1 volume
 for
> the redo logs.

        REDO logs should not be RAID5 and should be mirrored.

> 3. To reduce the possibility of disk contention, the archive log
 files
> should be separate from both the db & the active redo log files. To
> keep the number of disks down to a minimum, under what circumstances
 can
> we keep the archive log files on the same disk as the active log files
 

> (assuming we have RAID1 for fault tolerance)?

        If you are concerned enough about performance to want to keep your archive logs on a separate drive from your REDO logs DO NOT USE RAID5 FOR YOUR DATABASE!         I have found that if you have sufficient REDO log groups and files you can keep the REDOS and the archives on the same drive without incurring a significant performance penalty. Again, if you are going for maximum performance on the box separate them.

> 4. How does one size the redo log files? When do we use, say, 3
 10MB
> files vs. 3 100MB files (with ARCHIVELOG on)?

        Experiment. Your goal is to make the logs large enough so that log switches occur no more than three times an hour. You then want enough log groups and members so that when switching does occur there will be no waits. Note that these are ideals that may not be attainable given the distribution of your workload.

> Thank you.
 

> Sybrand Bakker wrote:
> >
> > david spaisman <david.spaisman_at_compaq.com> wrote in message
> > news:8c303v$6hm$1_at_mailint03.im.hou.compaq.com...
> > > Sybrand:
> > >
> > > Thanks for quick and helpful reply.
> > >
> > > I had thought that mirroring the redo logs on separate disks would
 

avoid
> > > problems if one memeber of a redo log is corrupted. How would I av
oid
this
> > > problem : in my configuration, I beleive I should place the redo
 log
> > > memebers on disk 1, 4 and disk5.
> > >
> > > Your reply on not putting critical file on RAID 5 devices would l
ead
me
> > to
> > > have in my configuration(not the actual drive letters to be used):
 

> > >
> > > Drive a) if this is to be raid 5, I should keep my system tablespa
ce
on
> > disk
> > > d or disk e. If I have disk a set up for RAId 1 I would be ok?
> > >
> > > If disk a,b c, d and e or disk b,c ,d,e are raid 5 enabled, where
 do I
> > > spread out the control files-- on which drives?
> > >
> > > Thanks again for the very helpful response. Unfortunately, transla
 ting
 dutch
> > > would be an eternity for me.
> > >
> > > David Spaisman
> > >
> > >
> > > Sybrand Bakker wrote in message
> > > <954531184.21961.0.pluto.d4ee154e_at_news.demon.nl>...
> > > >Answers embedded
> > > >david spaisman <david.spaisman_at_compaq.com> wrote in message
> > > >news:8c2sr8$55r$1_at_mailint03.im.hou.compaq.com...
> > > >> Hello:
> > > >>
> > > >> I am working on Oracle 8.0.5 application on NT 4.0. We are in t
 he
 process
 of
> > > >> setting up a user acceptance server and a development server an
 d have
 the
> > > >> luxury of having as many disk servers as needed on each respect
 ive
> > > >> server(within reason).
> > > >>
> > > >> I am thinking of going with a configuration of drives consist
 ing of
 the
> > > >> following:
> > > >>
> > > >> a) Oracle executibles, redo log group members, control file, sy
 stem
> > > >> tablespace
> > > >> b) data files, user files, control file, redo log members
> > > >> c) index files, control file, redo log members
> > > >> d) rollback segments, export files, backup files
> > > >> e) archive log files.
> > > >>
> > > >> Hopefully this configuration will be with physically separate d
 rives
 and
> > > >> more than one controller If these are logically partitioned dri
 ves, I
> > > >> believe it will still depend on how many physical drives and
 controllers
> > > >> are involved. THanks.
> > > >>
> > > >> David Spaisman
> > > >>
> > > >>
> > > >> However, I have been told that RAID 5 will reduce or contradi
 ct the
> > > >> ebenfits purportedly gained from the multiple disk drive/OFA
 configuration.
> > > >>
> > > >> 1) Has any one found this to be true?
> > > >>
> > > >Yes, basically you really can't distribute your data as all your

logical
> > > >volumes are spread out on several physical volumes. If you dedica
 te 1
 RAID
> > > >disk to indexes, your indexes will still show everywhere.
> > > >
> > > >
> > > >> 2) Will the benefit of RAID 5 -- faster reads versus slower wr
ites

--

> > > for
> > > >a
> > > >> transactonal database still apply?
> > > >>
> > > >No, RAID 5 will hurt performance and cause bottlenecks especially
for files
> > > >being sequentially written only, like redo log files. Your setup
with
> > > >redolog files on disks with tablespaces is likely to result in
performance
> > > >hits.
> > > >> 3) Has any one seen Oracle position on the value OFA versus t
he benefit of
> > > >> RAID 5?
> > > >>
> > > >No, though the consensus in this group is : use a combination of
RAID0+1
> > > and
> > > >RAID-5, do NOT place critical files on RAID-5 devices.
> > > >A recent article in the Dutch Oracle Magazine Optimize summarizes
as
> > > >follows.
> > > >If your database has less than 50 users and/or less than 250 OLTP

> > > >transactions per minute there should be no objection against RAID
5. If
> > one
> > > >of these parameters is exceeded and/or you are running more datab
ases on
> > > one
> > > >server, you should consider using other disks.
> > > >The article discusses heavy OLTP environments, I'm not sure wheth
er you
> > > need
> > > >that, and it will be a hell of a lot of work to translate from Du
tch to
> > > >English.
> > > >
> > > >Hth,
> > > >
> > > >Sybrand Bakker, Oracle DBA
> > > >
> > > >> 4) Any other information concerning this situation will be grea
tly
> > > >> appreciated.
> > > >>
> > > >> Thanks.
> > > >>
> > > >> David Spaisman
> > > >>
> > > >>
> > > >>
> > > >>
> > > >
> > > >
> > >
> > >
> >
> > With 5 disks, I would probably make sure two are non-Raid or RAID0+1
.
> > According to the article RAID5 should get
> > software
> > exports
> > hot backup
> > archive logs
> > The rest is on RAID0+1 or RAID10 or non-RAID
> >
> > With respect to redo log I have always believed you should use eithe
r hw
> > mirroring (RAID) or sw mirroring (multiple members), but not BOTH. I
t won't
> > help you either, if one of the disks of that stripe set goes down, t
he
> > database will probably detect inconsistencies and crash.
> > In your config I wouldn't definitely not use 3 log file members, onl
y 2.
> > Given the time of the day here (22:16) I would probably not endeavou
r to
> > come up with a different config.
> >
> > Hth,
> >
> > Sybrand Bakker, Oracle DBA

> --
> Buzz Huse E-Mail: mailto:buzzhuse_at_flash.net
> Euless, Texas, USA Homepage:
http://www.flash.net/~buzzhuse/
> "These opinions/comments are entirely my own and no one else's."
Received on Sun Apr 16 2000 - 00:00:00 CDT

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US