Re: Windows2003 Server Config for Oracle 10g

On Feb 8, 5:11 pm, "Charles Hooper" <hooperc2..._at_yahoo.com> wrote:
> On Feb 8, 1:29 pm, "red_val..._at_yahoo.com" <red_val..._at_yahoo.com>
> wrote:
>
>
>
> > On Feb 8, 10:59 am, "Charles Hooper" <hooperc2..._at_yahoo.com> wrote:
> > > On Feb 7, 4:41 pm, "red_val..._at_yahoo.com" <red_val..._at_yahoo.com>
> > > wrote:
> > > > The PE2900 is configured with 8 GB memory; 2x73GB mirrored root SAS
> > > > drives on channel0; and 8x146 GB SAS drives in 2 RAID 10 arrays with
> > > > each stripe set of 4 drives on separate channel of the I/O card.
>
> > > > The root disks will be logical C: drive, others logical D: drive.
> > > Is this the 32 bit version of Windows 2003, or the 64 bit version?
>
> > > I am a bit curious why you broke the eight drives into two RAID 10
> > > arrays of four drives each, rather than leaving the eight drives as a
> > > single RAID 10 array.
>
> > > Take a look through the following, and read the "Performance Tuning
> > > Guide" guide cover to cover before deciding how to set up the server.
> > > "Performance Tuning Guide"
> > > http://download-east.oracle.com/docs/cd/B19306_01/server.102/
> > > b14211.pdf
>
> > > Charles Hooper
> > > PC Support Specialist
> > > K&M Machine-Fabricating, Inc.
>
> > My intention is to have each _stripe_ set on a single channel for as
> > near to simultaneous mirroring across channels as possible, like so:
>
> > StripeSet1 StripeSet2
> > ************ ************
> > Channel1 Channel2
> > -------------- ----------------
> > disk01 =====> mirrordisk11
> > disk02 =====> mirrordisk12
> > disk03 =====> mirrordisk13
> > disk04 =====> mirrordisk14
>
> > Channel 0 supports the root disk mirrored pair.
>
> > Does this make sense?
>
> > Now I don't know whether the Dell PERC 5/i RAID card can support this
> > since the Dell online documentation is not sufficiently detailed, but
> > this seems to be a (conceptual) way of optimally configuring the bulk
> > storage. If the card doesn't support, this then it'll have to be one
> > big RAID10 array.
>
> > Hey -- can I further logically slice the devices (either through an O/
> > S utility or card vendor software or 3rd-party tool) so that the D:
> > drive comprises the first two mirrored pairs and the second two
> > mirrored pairs are the E: drive? I just don't know.
>
> > My goal here is to do it right from the start (that last not a
> > political statement), and avoid causing myself the angst I've
> > experienced at nearly every IT shop I've walked into where I've found
> > that things are not configured well, or even correctly, usually due to
> > ignorance (or neglect) on the part of those responsible for the
> > systems.
>
> > But it's hard to know what you don't know, and even that you don't
> > know, hence my original post.- Hide quoted text -
>
> > - Show quoted text -
>
> I believe that I understand what you are trying to accomplish with the
> RAID 10 arrangement. It has been a while since I read through Dell's
> RAID controller documentation. I believe that all drives in the same
> RAID array must be present on the same controller, and likely also on
> the same channel of the controller.
>
> If the 8 drives are in a single RAID 10 array, with a 64KB stripe
> depth (64KB * 4 = stripe width), and the server is told to write out a
> file that is slightly over 512KB in size, the file may be split among
> the hard drives by the RAID controller similar to this:
> DRIVE 1: 0KB - 63.99KB DRIVE5: 0KB - 63.99KB (Mirror)
> DRIVE 2: 64KB - 127.99KB DRIVE6: 64KB - 127.99KB (Mirror)
> DRIVE 3: 128KB - 191.99KB DRIVE7: 128KB - 191.99KB (Mirror)
> DRIVE 4: 192KB - 255.99KB DRIVE8: 192KB - 255.99KB (Mirror)
> DRIVE 1: 256KB - 319.99KB DRIVE5: 256KB - 319.99KB (Mirror)
> DRIVE 2: 320KB - 383.99KB DRIVE6: 320KB - 383.99KB (Mirror)
> DRIVE 3: 384KB - 447.99KB DRIVE7: 384KB - 447.99KB (Mirror)
> DRIVE 4: 448KB - 511.99KB DRIVE8: 448KB - 511.99KB (Mirror)
> DRIVE 1: 512KB - 512.99KB DRIVE5: 512KB - 512.99KB (Mirror)
>
> All eight hard drives were were engaged twice in the write operation,
> and drives 1 and 5 were engaged twice. There is a document on
> Oracle's website that describes Oracle's SAME method and discusses
> performance justifications (add this document to your must read list),
> which concludes that the optimal stripe size is usually about 1MB - it
> was not clear to me whether stripe size in this case refers to the
> stipe depth (64KB in my example), or the stripe width (256KB in my
> example). As best that I could tell, Dell's RAID arrays support a
> maximum stripe depth of 64KB.
>
> Now the question, can you break the eight drive RAID 10 array into
> multiple drive letters? The question that I have is why? Doing so
> may hurt performance, as the drive heads would need to travel greater
> distances from the starting cylinder of one partition to the starting
> cylinder of the next partition as data is written to and read from the
> two partitions. Considering the size of the hard drives that you
> selected, if you leave the drives with a single partition, you should
> have minimal delay in track to track seek time.
>
> Dell may have partitioned your RAID 1 array into a C and D drive. The
> D drive may be ideal for the Flash Recovery Area, where archived redo
> logs may be stored.
>
> Charles Hooper
> PC Support Specialist
> K&M Machine-Fabricating, Inc.

Charles,

Thanks you ever so much for the terrific advice. This is precisely the response I was hoping for. Y'all saved the cost of 4 GB of unnecessary memory -- something rather rudimentary which I should have noticed in the first place.

What I'm trying to do with the DOS -- oops, meant Windows -- logical drives is enable data to be optimally partitioned vis-a-vis the RAID array. Since database performance is intimately linked to placement of most heavily used data structures on disk (databases/tablespaces/ tables) I'm trying to discern how -- and how best -- to slice and dice the bulk storage on an unfamiliar O/S with foreign DBMS. I'm really trying to avoid misconfiguring the host, as I've seen so many other places. The assumption always seems to be that the technology -- the RAID array, or filesystem, or some other nebulous black box -- will alleviate the SA's or DBA's of the effort to correctly arrange the available resources for optimal performance. It only seems to perpetuate ignorance.

I'm all ears for any other good scoop -- websites, (paper) references, anecdotes, experiences. Received on Fri Feb 09 2007 - 13:03:40 CST