Re: Questions on RAID-0 and RAID-0+1 Consideration

Hi Qun,

It is not clear from your post whether you intent to build 2 volumes striped across 4 all disks, or 2 volumes striped across 2 disks each. The former would be a bad idea, because you thrash the disk heads back and forth over the disks for every log write, because Oracle will write to both log members together - so I assume you're wont do that. But the alternative is not wonderful either, because Oracle needs at least 2 log groups, and so your archiver will content with your log writer after a log switch. So better not to stripe at all and follow Oracle's "archiver best practice" recommendation - 2 groups of 2 members, on 4 dedicated disk, with the archive destination on a 5th. Ideally the log files should be built first, to put them on the outside edge of the disk, for best performance. Then if you have disk to burn you can start to improve this picture. The first improvement would be to use 3 groups, rather than two, to protect yourself against slow archival (especially if your archive destination is on slower storage, as it should be). I think this is the 6 disk solution you are considering. The second improvement is to multiply the number of disk used for logs by N and stripe them. Under Solaris the optimal interlace (stripe element size) for redo logs is 64K/N.

On question B, I'd definitely go for the multiple mount points.

HTH, Steve Adams

On Mon, 19 May 1997 13:49:21 -0700, Qun Li <qun.li_at_DELETE.attws.com> wrote:

>Hi, all:
>
>We have a new SUN Ultra Server with 30 2GB external disks. This is for
>an OLTP Oracle 7.1.6 database. We like to use mix RAID-0 and RAID-0+1
>for all Oracle datafiles. Here is our initial setup:
>
>A. 4 disks use RAID-0 for redo logs. This will have 2 mount points: one
>for three redo groups, export dump files and the other one is for the
>mirroring members of redo logs. Since RAID-0 is stripping only so we can
>have maximum write throughput and second mount point will serve as
>mirror to against disk failure.
>
>Question is, would it be better if we use 6 non RAID disks and put three
>redo logs and their three mirror members into separated disk?
>
>
>B. 26 disks use RAID-0+1 for rest datafiles, archives and backup
>directory. We will have 5 mount points:
>
>1. Archive and backup directory
>2. SYSTEM, TOOLS, and Control file 1
>3. RBS tablespaces, and control file 2
>4. DATA tablespaces, and Control file 3
>5. INDEXES, TEMP, and USERS tablespaces.
>
>Question here is, is it better if we use one mount point for all or use
>more mount points to slicing the tablespaces into more separated file
>system?
>
>Thanks in advance!
>
>Qun
>
>qun.li_at_attws.com
Received on Sat May 24 1997 - 00:00:00 CDT