Yes, mainframes ARE more expensive than racked servers.  And
slower than most current Intel or Sun solutions - for single
applications.  There are no TCO studies published showing
otherwise that I know of AFTER an implementation is complete -
I'm only guessing at the reasons.... Is Meta group right?

It's pretty hard to fault their recommendation:

>We recommend that organizations match the type of work being done and
>the service-level goals, such as availability and scalability, to the
>appropriate computing platform. A key consideration should also be
>existing operations process maturity and staff skill levels. However,
>workloads that easily scale out across multiple commodity servers
>should not be centralized onto more expensive platforms simply because
>of skill and process issues. In this case, the right answer is to
>improve the skills and processes associated with running scale-out
>environments.


But to sell Linux on s/390, we have our ways. "Racked servers"
very often have very low utilization, either because the
platform can not handle high utilization, or the application
just doesn't utilize the hardware.  On the s/390 side, we expect
the processor to reach 80% (plus) utilization as the article
indicated.  Thus moving 20 servers that have 3-5% utilization to
s/390 may be less expensive and probably doable.

The mistakes in moving applications to Linux on s/390 are to
take high utilization applications and move them to s/390.  Those
fail from a TCO Perspective just because with the rather slow
z/900 processor, the resource requirement of the application is
just too high.  For this reason, I believe any Linux TCO analysis of
LPAR or dedicated s/390 will ALWAYS fail - just because what
makes the s/390 successful is a high level of multiprogramming
with z/VM - on a highly available system.  Comparing mip to mip,
storage cost to storage cost, s/390 will ALWAYS cost
significantly more. But utilized mip to mip is different animal.

To do a successful TCO showing s/390 to be less expensive
means you have to measure an application (some of you saw
this coming) on the racked server, port it to z/VM, and
measure it there.  With these numbers, you now have an
important part of a valid methodology to choose servers to move
to the s/390 that have a cost benefit.  If you already have z/VM
in house, this is much easier.  If you don't, it's pretty hard to
make this analysis in your own environment.  If you need the
tools to help you with this study, I can help...

At SHARE, I hope to have 1 or 2 of these comparisons ready
to present. Watch this space.  And I'm always looking for
more installations interested in the performance comparisons
and analysis....

>
>We are in the eternal struggle - trying to get Management to try
>LINUX on the zSeries.  However, we are continually faced with the
>costs of Mainframe against Unix and Windows Servers.
>
>Does the note below make sense or is there a counter argument.
>
>Look forward to help form the team.
>
>Paul Tormey
>LAN Services
>Standard Bank of South Africa.
>
>011 636 4103
>083 252 5292
>[EMAIL PROTECTED]
>







"If you can't measure it, I'm Just NOT interested!"(tm)

/************************************************************/
Barton Robinson - CBW     Internet: [EMAIL PROTECTED]
Velocity Software, Inc    Mailing Address:
 196-D Castro Street       P.O. Box 390640
 Mountain View, CA 94041   Mountain View, CA 94039-0640

VM Performance Hotline:   650-964-8867
Fax: 650-964-9012         Web Page:  WWW.VELOCITY-SOFTWARE.COM
/************************************************************/

Reply via email to