I don't know what your costs are....
I was under the impression that Websphere is a $25K per server (or
server engine) expense. $25K times 200 servers is a cool $5 million.
Add on the NT licenses, any database licenses and or any other package
that is on all servers and that is "BIG" bucks. That vs the cost of
Websphere for the mainframe on an 8-way would be $200K. That is a lot
of money left over. (I don't know if an 8-way would handle the load, it
was just a stake in the ground for discussion purposes.) Add $360K for
the VM license and you still have a lot left over from the $5 million.
Then add in the hardware, networking, maintenance of 200 platforms,
backup/recovery/disaster recovery, Operational cost, electrical,
cooling...yadda...yadda...yadda.. A mainframe solution might end up
being very cheap in comparison.
Especially true if the workloads of all the applications don't normally
peak at the same time. Also, development/testing images are almost free
in the mainframe side.
I'm sure that you are a big enough playor for IBM to come out to see
how well of a fit this would be. Just the brute size of this, plays
into the traditional mainframe strengths.
Tom Duerbusch
THD Consulting
>>> [EMAIL PROTECTED] 09/15/03 01:20PM >>>
Having received a response from Mark and Tom I would ask the following
question, what is the feeling among the group about consolidating a
couple
hundred Unix servers running WebSphere and databases onto Linux/390
under
z/VM. I guess to better phrase it based on Tom's response, would
WebSphere application servers be classified as transactional in the
opinion of the group? I of course understand that it ultimately
depends
on the application written but Java in general, what would be the
implementation most would lean towards? (Linux on Intel or Linux/390
for
z/VM)
Thank you, again the opinions and feed back are appreciated.
Eric Sammons
(804)697-3925
FRIT - Infrastructure Engineering
Tom Duerbusch <[EMAIL PROTECTED]>
Sent by: Linux on 390 Port <[EMAIL PROTECTED]>
09/15/2003 12:06 PM
Please respond to Linux on 390 Port
To: [EMAIL PROTECTED]
cc:
Subject: Re: Is Linux on z/VM suitable for "large
workloads"
Well, yes and no and yes and no and it depends.
For a long time, one of the benefits of Linux/390 on the mainframe was
the consolidation of many small servers onto fewer engines. This is a
cost thing. A license for a server is the same for Intel as the
manframe. And with VM (ok I will even include LPARS), you can have
multiple "servers" with much lessor license cost.
For the first few years, we were warned about putting large workloads
(read CPU intensive) on the mainframe. The MIPS are still cheaper on
Intel.
However, over the last 9 months, there have been, at least, two
discussions of shops putting very large Linux workloads on the
mainframe. One shop had 9 IFL engines running 7 Linux images. To me,
that sounds like a large work load, not server consolidation. Another
shop, about 6 weeks ago, talked about a large Linux/390 application,
without VM taking a large number of IFL engines. By the tone of the
discussion, it was the best I/O delievery platform available for their
application.
So, based on the above two cases, I nolonger reject the notion of
large
workloads under Linux/390. I now believe that, when it comes up, it
is
a subject to look at. Perhaps, in certain cases, it is cost
justifiable. Perhaps it is a more expensive solution, but the best
solution.
Being a VM bigot, both these cases left VM out in the cold. I
understand why. $45K per engine plus $11K/engine/year which is good
for
many "little" workloads, but with only a few engines, LPAR seems to be
worth it. (Ok it's basically free)
And with IBM's "capacity on demand", that adds a lot of cost savings
if
you have a highly variable workload.
Tom Duerbusch
THD Consulting
>>> [EMAIL PROTECTED] 09/15/03 07:19AM >>>
I have been under the impression for some time that zLinux, while a
terrific solution for some applications, may not be the best solution
or
fit for "large workloads". Certain colleagues of mine have begun
evaluating where zLinux might fit in our organization should we choose
to
go in that direction. They have listed Linux on Intel as suitable for
small workloads and zLinux as suitable for large workloads. In fact I
have been of the understanding that zLinux would be a solution for
many
guests installed under z/VM running small to medium workloads, i.e web
server farm. I keep thinking back to the Linux 2.6 kernel which
allows
for a larger number of CPUs, more memory and more disk to be
addressed.
In
addition I see Los Alamos implementing 2000+ AMD CPUs for almost 12
trillion operations per second of processing power (This information
from
ZDNet regarding Lightning, August 14, 2003). To me that sounds like a
large work load on an Intel type platform.
So my question is, to those of us using zLinux, what are we seeing it
implemented for and where is working best?
Thank you, This information will prove very helpful.
Eric Sammons
(804)697-3925
FRIT - Infrastructure Engineering
