Re: Oracle Databases, NTFS file systems and file system fragmentation ?

Hi Sybrand,
We've just have to make such a decision on raid 5 or not and it went as follows:

Sun E10k , 1 virtual server for our database with 4 cpu, 4G RAM. We had to choose the proper disk array. First our hardware and oracle "experts" have ordered a D1000 (no raid controller, 4x 36G disks). Of course this was canceled. I've insisted on buying a A1000 with 12 disks with a scsi raid controller (I would have made some raid1s and redo logs on 2 single disks) but the sun people have strongly recommended a T3 disk array on fiber channel with 9x 18G disks with one big raid5. After hours of conversations I've decided to call the oracle support (in austria) and they told me that I wouldn't have any performance problems at all, even the contrary. I would have problems only if I run in archive mode and we run in noarchive. So i had to give up and we've ordered the T3. I was the only one who didn't recommend this T3. The sun representative has told us that the disks run with 100MB/s theoretical and about 90MB/s practical throughput (every single disk). The fiber channel run also 100MB. Do you know whether it's true? The redologs would be also on this raid5. Our db is about 56 GB of netto tablespaces , currently running on NT 8.1.6 PP200, 256MB and about 14 disks (very old, 14ms access time, 5400rps) on some raid1 and some single disks. There are jobs running every day processing 2.5-20 Mio recs per job (depending on the job). With the new machine we want to achieve a performance win of min 300% (instead of 17 hours duration, 5 and so on) Do you think this "wonder" disks and disk arrays from sun would be a bottleneck. Can you give me a tip how to determine this bottleneck in a short time after the machine is configured (in order to push a change of this array before we go into production)
Thank you in advance
Alex

Sybrand Bakker wrote:

> If that raid disk system is raid-5, get rid of the raid-5.
> Raid-5 has CRC for every write, which is a major bottleneck for all writes
> to redolog.
> If that is 100G data on 1 single raid-array, on one disk-controller, I'm not
> going to tune anything: the raid-array is the source of the problem as the
> recommendation from Oracle and other sources, to distribute I/O amongst
> different individual disks are clear enough.
> I'm now at a shop where I have one raid-array, consisting of 3 disks,
> configured as a stripe set, configured as three *logical* disks, and the
> performance is a disaster (and of course they didn't consult us when buying
> that server). Also, when you loose one disk in that array, as you can't
> determine which file is on which disk, you basically will loose your
> database.
> I'm not sure how much time it would take to restore 100G, but I wouldn't be
> prepared to live with your situation.
> In my experience Oracle doesn't suffer from fragmentation at all, provided
> of course, you don't are continually adding new files to your database.
>
> However, talking about application problems: did you our your firm do
> everything to isolate inefficient sql?
> Usually it's not the hardware, but the application, and most developers just
> don't want to admit that.
>
> Hth,
>
> Sybrand Bakker, Oracle DBA
>
> "Al Dykes" <adykes_at_panix.com> wrote in message
> news:8qgm19$png$1_at_panix2.panix.com...
> > I'm not an oracle admin but I'm getting involved with an NT Oracle
> > (8.1.5) server with application performance problems. I'd like to
> > hear about possible affects of ntfs file system fragmentation and what
> > production shops do about it, if/when it happens. We have about 100GB
> > of data on raid disk system.
> >
> >
> > Thanks.
> >
> >
> >
> >
> > --
> > Al Dykes
> > -----------
> > adykes_at_panix.com
> >
  Received on Mon Sep 25 2000 - 12:13:39 CDT