Re: Fundamental RAID ISSUES

On Fri, 18 Jun 1999 10:26:29 GMT, coakleyj_at_hotmail.com wrote:

>A couple of fundamental questions regarding RAID 5
 configurations for an
>Oracle OLTP application, involving

  60% query, 40% writes approximately (NT
>4)-

1. In general, which is the better configuration (and why) - Six 9GB

>10K RPM disks or Three 18GB 10K RPM disks - both configured
 as RAID 5. (I
>presume the more disks the better to reduce disk I/O?).
 Is there an upper
>limit to the number of disks that can be
 configured as one RAID set
>(before management overhead

 becomes too large)?

I'd select the 6 disk option.
With RAID5 the redundancy data always consumes a whole disk's worth of data. So with the 9GB disks you'd end up with 45GB of useable space. With the 18GB disks you'd have only 36GB. There are other performance reasons why more disks are ALWAYS better than fewer disks.

2. In terms of the amount
>of disk space "used-up" on the

 RAID parity - how does this differ in the
>above configurations?

 How is this calculated? - It seems to be a function
>of the number and size of the disks in the RAID set.
 Is there a formula?

See #1 above.

3.
>What are the implications of the "stripe width" when
 configuring the system
>- What's "normal" ?

 How does this interact with the O/S block size (if at
>all),

 and how do these settings interact with the Oracle block size
>setting? Which drives the Oracle block size -
 the stripe width or O/S block
>size?

4. Previous questions and answers on this SIG have suggested that
>RAID 5 decreases write performance

 whilst improving reads. The reason given
>(I think) was

that the write involved 2 writes really - one for the data

>and one for the parity. Does the same logic no
t apply to reads - i.e. with
>RAID 5 we need to perform

2 reads instead of one ? Hence reads are slower as
>well??

Can the system be configured to optimize reads over writes or vice
>versa?

With RAID5 one XOR (redundancy) data block is updated (written) for every user data block written. ALWAYS. This is just the nature of RAID5. There is no "overhead" or penalty with RAID5 read operations.

5. Finally, a licensing issue regarding NT4- Assuming we have 50
>concurrent Oracle users on an NT4 server. Given
 that the users are operating
>client-server and only

 need "ping" capability to the NT server (i.e. the
>end-users

are not set up as users on the NT server, nor do they  have access
>to the disk drives), do we need to

 licence 50 users of NT or not? If not,
>how many?

Regards
Coakleyj
>
>
>Sent via Deja.com http://www.deja.com/
>Share what you know. Learn what you don't.
Received on Fri Jun 18 1999 - 10:37:47 CDT