Re: DBA Experiences with Oracle and RAID 0+1

[Mogens N�rgaard]

Jared,

Good thought! We're running our annual Miracle Master Class (this year with Jonathan Lewis) this week and Cary Millsap (and his gang), Jonathan and James will arrive tomorrow (Monday), while Steve Adams, Lex de Haan, Stephan Haisley and a bunch of other guys will arrive on Tuesday. I'll keep this message and throw it up for discussion on the Oak Table (see Cary's wonderful article on www.Undskyld.Org)...

Jared Still wrote:

Mogens,

In regard to the number of spindles issue:  James Morle has some 
excellent discussion on that in 'Scaling Oracle 8i'.   ( I think it's that
book  )

When some of the newer larges drives are used in a given configuration,
they mabe be able to outperform older drives in a similar configuration with
a larger number of spindles.

I say 'older' since smaller drives usually aren't using the latest technology
and the newer ones have sufficiently higher throughput to match the 
capability of a larger number of drives in given configuration.

Food for thought, anyway.

Jared


On Sunday 20 January 2002 06:50, Mogens Nørgaard wrote:

Diego,

I agree with you 100% and didn't express myself correctly in my email.
The more spindles the better. What I meant to say was that you must
never buy disks by taking your total needed amount of space and divide
by the number of big disks you can get hold on :). It's the number of
IO's required by the disk system that matters, not the size...

Thanks for making this clear to everyone.

Mogens

Diego Cutrone wrote:

Mogens:
   Just let me disagree with you at only one point. According to my
experience, I think that the size of the disks in an array does matter
sometimes. It's not the same to have 24 9GB disks that to have only 3 of
73GB. You have 24 spindles againts 3, the first option (in a well
configured system of course) will give you better performance in
enviroments where you have a lot of concurrency and many users.
   However I think that what I've written above might not be correct (may
be it should be tested) if the 73GB outstands for a long way the 9GB disks
in terms of seek time and transfer rate.
   Take a look at an extract of Gaja's paper "Implementing RAID on
Oracle":

"5) Procure the smallest drive money can buy, keeping in mind scalability,
limits of the host
machine, the disk array and growth projections for the database. This is a
tough one these
days, with 18 GB drives considered as small drives.<
br>
6) Bigger and faster drives are not always better than smaller slower
drives, as the seek times
for larger and faster drives with larger form factors, may be more than
their smaller and
slower counterparts. This is not that big of an issue, if your drives
support a built-in track
buffer cache for storing an entire track's worth of data from read
request(s)."


HTH
Greeting
Diego Cutrone

----- Original Message -----
To: Multiple recipients of list ORACLE-L <[EMAIL PROTECTED]>
Sent: Tuesday, January 15, 2002 10:25 PM

Jon,

It's one of those "how many bags will I need in the supermarket?"
questions - it depends.

Consider:

- RAID 1+0 is much better than 0+1.
- Three disks is not much w.r.t. IO capability. If you have three
concurrent users you'll be OK :)
- Size doesn't matter (who cares if it's 10, 36 or 73 Gig disks? It's
the IO capabilitity that counts)
- I'm new to this list, so I don't know if this will work, but I've
attached a brilliant presentation by our old friend James Morle (check
out www.ScaleAbilities.com) regarding SAN, NAS and RAS (Random Acronym
Seminar).
- If you're only striping across three disks (is that really a SAN?)
just SAME (Stripe And Mirror Everything). It might not be good, but it's
simple.

Jon Behnke wrote:

We are in the process of setting up a SAN using RAID 0+1 for our

database.

In our current environment, we are able to separate our tables, indexes,
rollback segments, and archive logs on different disks.  On the SAN we

would

have six 73 gig disks on RAID 0+1 for a total of about 210 Gig of usable
space (3 disks worth of space).

Some white papers that I have read suggest attempting to separate the

data,

indexes, and rollback segments on separate RAID volumes, and others

simply

suggest that the performance boost of striping will supercede the

separation

of these items.

Can anyone offer any comments or suggestions?

Jon Behnke
Applications Development Manager
Industrial Electric Wire & Cable
Phone (262) 957-1147  Fax (262) 957-1647
[EMAIL PROTECTED]

----------------------------------------
Content-Type: text/html; charset="us-ascii"; name="Attachment: 1"
Content-Transfer-Encoding: 7bit
Content-Description: 
----------------------------------------