RE: New 270. -- MIDRANGE-L

Matthias,

I went back to some of your earlier posts.  You are replacing a 9402-400
(2131) with a 9406-270 (2431).  Your current system has six drives,
mirrored load source with a RAID set of four.  You are moving from a 20
CPW machine to a 465/30 CPW machine.  This new machine will fly on your
current workload.  Your users will be very impressed.  You've stated
that you have 20-30 users, I assume that this is total workstations, not
the number of people who are likely to hit the enter key on a
transaction at the same time.  Even with some normal batch throughput,
this is a very small workload, probably considerably smaller than the
interactive workload described in the Disk Arm Requirements document.

I agree with your hardware and software vendors.  Unless you are
planning on adding new workloads, either disk controller would probably
give you comparable performance.  Two of the significant differentiators
between the two are the size of write cache and the number of disks
which can be attached.  In your situation, it probably doesn't matter to
you whether you can attach 12 (#2763) or 18 (#4778) disk units.  The
#2763 has a write cache of 10 MB versus the 104 MB of the #4778.  It
probably doesn't matter to you whether you can compress your disk
storage or not, or whether you can attach an adaptive cache device.

The only significant variable between the controllers in your
configuration would be the size of the write cache.  If you think of how
many adds/updates are occurring on your system at any one time, do you
think you will exceed 10 MB?  It doesn't sound like it to me.

It sounds like money is an important factor in this purchase; I would go
with the lesser of the two controllers.  I don't think there would be a
noticeable difference between the two.

Regards,
Andy Nolen-Parkhouse

> Subject: Re: New 270.
>
> Hi,
>
> Thanks again for the continuing discussion.
> On the web I've found an interesting document called "iSeries Disk Arm
> Requirements based on Processor Model Performance"
> (to be found here:
> http://www-1.ibm.com/servers/eserver/iseries/perfmgmt/pdf/iSarmct.pdf
)
>
> It explains why the number of arms are an important factor and has
> tables and formulas to calculate diskarm requirements based on model,
> feature, disk type, controller type, etc.
>
> Given the 270 I'm looking at (270, 2431, v5r1, Raid-5, 6 4318 disks
and
> a 2763 Raid controller, I calculate 7.98 or 8 disks.
> If I was to replace the 2763 with a 4778 Controller the arm
requirement
> reduces to 6.098 or 6 disks.
>
> Problem solved I thought, different controller.
> However, the hardware and software vendors are quite insistent that a
> 4778 would be total overkill, that performance using a 2763 would be
> fine.
>
> I like the advice because the 2763 costs only half as much as the
4778.
> But I'm not all that sure it's the right advice from a performance
point
> of view. On the other hand, it can be safely assumed that, given the
> more capable processor, on average our CPU use would be well below
70%.
>
> Any comments would be most welcome.
> Best regards,
> Matthias