Discussion:
MIPS. Still Used? Why? (Was: 2086 230 & 240 MIP rating)
(too old to reply)
Al Sherkow
2004-11-12 19:58:43 UTC
Permalink
IBM really changed this discussion when the z990 had one Announced MSU
size, and a smaller MSU number for Software Pricing.

With their "mainframe charter" (22 August 2003) IBM announced on the web
what they called "Improved Software Pricing on z990 Servers", with an
effective date of 10Oct2003. They created what I named "Software Pricing
MSUs" (SWPMSUS), though none of IBM's material uses that term. These MSUs
did not change the "announced MSU Capacity of the z990", but the Software
Pricing MSUs are approximately 10% smaller than the "Announced MSUs" of the
z990s. This decreased the cost of all IBM software that was charged on a
capacity basis; including PSLC, WLC, and ULC. IBM also was encouraging ISVs
to adopt these Software Pricing MSUs for any of their capacity based
pricing plans. You can read about this at the following link:

http://ibm.com/zseries/announce/charter/pricing.html


and the key information is in this paragraph:


"Improved Software Pricing on z990 Servers. Superior performance and
technology within the z990 has allowed IBM to provide improved software
prices for key IBM zSeries operating system and middleware software
products. The superior technology includes superscalar processing, native
"LPAR-only" mode of operation with the capability to support 30 LPARs,
support for multiple logical channel subsystems within a single system
footprint, and complete support of z/Architecture. While the announced z990
MSU values remain unchanged, IBM is pricing the designated software on z990
at lower MSU values than those previously announced. In order to implement
this IBM software pricing initiative on z990, IBM intends to lower the MSU
values generated by the z990 microcode by approximately 10 percent."

I wrote about this on my website on 23Aug2003 and again on 17Sep2003
(http://www.sherkow.com/updates/aug2003.html). IBM's actual announcement
letter was 103-280 dated October 7, 2003 "IBM enhances the IBM zSeries 990
family of servers". By changing the hardware microcode, any vendor that
asks for the capacity retrieves the SWPMSU numbers.

MSUs have all the same traditional problems as MIPS, with one major
difference. IBM publishes MSUs
(http://ibm.com/zseries/library/swpriceinfo/hardware.html) and IBM never
published MIPS; they have been doing relative capacity for years with LSPR.
It is consultants and IBM-Watchers (Gartner Group, Meta Group, Cheryl) that
publish MIP tables. IBM only wants you to use LSPR, but your management
probably wants a single number. MSUs are a single number for each machine
model. The LSPR provides information that matches our collective
experience, i.e. "not all workloads run with the exact same performance on
a given machine model. This is partly why the people who sell MIPS tables
usually provide a range of performance. I do not have Cheryl's or Gartner's
tables in front of me, though I've seen them in the past and they included
a range of performance. I believe they do not do their own testing, their
ranges are based on the LSPR, anecdotal information, and experiences from
their clients and customers. The old-timers will remember that years ago
large organizations actually "benchmarked" machines. I don't think any
company does this anymore except for the vendors. Benchmarking is too
expensive. Even IBM does not test each model relative to each other model,
they compare a representative sample so customers can use the LSPR to
compare various machines and models. Gary King has an excellent
presentation on MIPS, MSUs and LSPR. His long experience in capacity
planning and working for IBM is nicely pulled together in this session.
Don't just read the session material, but attend and listen to Gary if you
have the chance, as I did last week at the z/OS Expo conference.

In my presentations at Share, z/OS Expo, CMG, etc., I have been telling
people to swing their management thinking over to MSUs. If you are not
using MSUs you should start to do so. One step may be to put MIPS on the
left vertical axis and MSUs on the right axis of your capacity and
performance charts. MIPS for the historical reference they provide and MSUs
because that is now how the software pricing and the hardware are being
sized. This is especially relevant as the software costs typically
represent a larger percentage of the upgrade costs versus the hardware
costs. You might just as well report the SWPMSU number.

With the z890s IBM only announced the software pricing MSUs. If you look at
the SRM constants (SMF72ADJ) and crank through some arithmetic, you'll see
the SWPMSUs are about 10% less than the HWMSUs calculated from SMF72ADJ. I
expect that any future announcements from IBM will only be Software Pricing
MSUs. Also, IBM can again change how much "service" is delivered with
Software Pricing MSUs relative to the HWMSUs we can calculate from the SMF
constants.

I'm a mainframe proponent and proud of it. The difference between SWPMSUs
and HWMSUs is important to systems programmers, performance analysts and
capacity planners. With sub-capacity pricing there is now an opportunity
for the technical staff to have a direct impact on costs in the data
center. Tuning, besides improving performance and reducing resource usage,
can now more directly influence software charges. I have clients and
licensees saving money on their software charges month after month with
sub-capacity pricing.

Al Sherkow, I/S Management Strategies, Ltd.
Consulting on Capacity Planning, Performance Tuning,
WLC, LPARs, IRD, MSUs and LCS Software
Voice: +1 414 332-3062 Web: www.sherkow.com




----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to ***@bama.ua.edu with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
McKown, John
2004-11-12 20:11:14 UTC
Permalink
-----Original Message-----
From: IBM Mainframe Discussion List
Sent: Friday, November 12, 2004 1:59 PM
Subject: MIPS. Still Used? Why? (Was: 2086 230 & 240 MIP rating)
<snip>
"Improved Software Pricing on z990 Servers. Superior performance and
technology within the z990 has allowed IBM to provide
improved software
prices for key IBM zSeries operating system and middleware software
products. The superior technology includes superscalar
processing, native
"LPAR-only" mode of operation with the capability to support 30 LPARs,
support for multiple logical channel subsystems within a single system
footprint, and complete support of z/Architecture. While the
announced z990
MSU values remain unchanged, IBM is pricing the designated
software on z990
at lower MSU values than those previously announced. In order
to implement
this IBM software pricing initiative on z990, IBM intends to
lower the MSU
values generated by the z990 microcode by approximately 10 percent."
<cynicism>

Translation: Software cost is causing the zSeries to die. So we're going
to reduce our margins on software and hope that ISVs follow suit so that
the zSeries does not go the way of the other mainframes that we drove
out of that business.

</cynicism>

<snip>


--
John McKown
Senior Systems Programmer
UICI Insurance Center
Information Technology

This message (including any attachments) contains confidential
information intended for a specific individual and purpose, and its'
content is protected by law. If you are not the intended recipient, you
should delete this message and are hereby notified that any disclosure,
copying, or distribution of this transmission, or taking any action
based on it, is strictly prohibited.

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to ***@bama.ua.edu with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
Eric Bielefeld
2004-11-12 21:44:33 UTC
Permalink
Hi Al,

IBM keeps making the customer jump through hoops to save money on
software. Why don't they just cut their software charges. I can see
giving a break on the software with new hardware, but just make it a lower
rate on the z890 or z990. Why make customers waste sysprog time just to
get their costs down. I know they did lower the costs about 10% with the
z/890 & 990, but z/OS is still way way more than any other operating system.

I listened to Rick Ralston's presentation at MCMG in Chicago last month on
saving money on IBM software costs. It was very interesting, but it just
seemed that IBM could just cut their charges more give and not make you go
to such extreems to save money.

I guess IBM's high software pricing upsets me because I'm sure that the
high price of z/OS is a big factor in my job going away. I'm sure that's
true of many other posters on IBM-Main at one time or another. Yeah, I
know that IBM can't drastically cut their prices, but as more and more low
to mid size customers leave the platform, I wonder!

Eric Bielefeld
P&H Mining Equipment
414-671-7849
Post by Al Sherkow
I'm a mainframe proponent and proud of it. The difference between SWPMSUs
and HWMSUs is important to systems programmers, performance analysts and
capacity planners. With sub-capacity pricing there is now an opportunity
for the technical staff to have a direct impact on costs in the data
center. Tuning, besides improving performance and reducing resource usage,
can now more directly influence software charges. I have clients and
licensees saving money on their software charges month after month with
sub-capacity pricing.
Al Sherkow, I/S Management Strategies, Ltd.
Consulting on Capacity Planning, Performance Tuning,
WLC, LPARs, IRD, MSUs and LCS Software
Voice: +1 414 332-3062 Web: www.sherkow.com
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to ***@bama.ua.edu with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
Ted MacNEIL
2004-11-12 22:24:43 UTC
Permalink
..
IBM really changed this discussion when the z990 had one Announced MSU size, and a smaller MSU number for Software Pricing.
..

I wonder why they didn't just drop the price per MSU by 10%.


***@bell.blackberry.net

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to ***@bama.ua.edu with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
Phil Payne
2004-11-17 13:58:52 UTC
Permalink
Post by Ted MacNEIL
...
IBM really changed this discussion when the z990 had one Announced MSU size, and a smaller
MSU number for Software Pricing.
Post by Ted MacNEIL
...
I wonder why they didn't just drop the price per MSU by 10%.
a) Some customers have paid up front and received a discount. Recalculating their liabilities
would require recovering the outstanding net present value, fiddling with it, and possibly
returning cash from the deferred revenue account. While IBM has always been exemplary with
its accounting, you can bet someone would have a look. What with local sales taxes around the
world, returning money is a nightmare. Interest on the tax component of prepaid charges held
in deferred revenue? UGH!!

b) IBM wants to sell systems. Making a MIPS or an MSU cheaper on new hardware helps to sell
it.

c) The discount is not constant across products or even models.

d) The hit would be 10% of software revenue (pure margin) with immediate effect right across
the installed base. Justifiable to the shareholders only if the z/OS market were elastic
enough to make up that 10% within a quarter or two. Which it isn't.

e) 10% isn't anything like enough to correct the malpositioning of z/OS relative to other
systems. It was merely to compensate for the z990's underperformance. 90% is a better
figure, which is the area of z/OS.e and NALC.

--
Phil Payne
http://www.isham-research.com

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to ***@bama.ua.edu with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
Ted MacNEIL
2004-11-12 22:38:40 UTC
Permalink
..
IBM publishes MSUs and IBM never
published MIPS; they have been doing relative capacity for years with LSPR.
..

IBM Canada always talked MIPS when I was at ScotiaBank. I still have a chart that they gave me that had a 3090-200E at a relative capacity of 32 Units. I was amazed at that 'coincidence' (8-{]}

LSPR has always been a bit suspect. The z/990 more so!


***@bell.blackberry.net

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to ***@bama.ua.edu with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
R.S.
2004-11-15 11:23:22 UTC
Permalink
Post by Ted MacNEIL
...
IBM publishes MSUs and IBM never
published MIPS; they have been doing relative capacity for years with LSPR.
...
IBM Canada always talked MIPS when I was at ScotiaBank. I still have a chart that they gave me that had a 3090-200E at a relative capacity of 32 Units. I was amazed at that 'coincidence' (8-{]}
Not only IBM Canada. I have a chart on the wall, printed by IBM (US I
guess), it's titled "Optimized for serious e-business", contains bars
and numbers - *in MIPS*.
The document number is G511-4172.

--
Radoslaw Skorupka
Lodz, Poland

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to ***@bama.ua.edu with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
Phil Payne
2004-11-17 13:58:29 UTC
Permalink
Post by Al Sherkow
IBM really changed this discussion when the z990 had one Announced MSU
size, and a smaller MSU number for Software Pricing.
It started with the z800. Until then, IBM's guidance on MIPS indicated roughly 6.36 or so
MIPS per MSU. The z800 was closer to 5.90 MIPS per MSU.
Post by Al Sherkow
"Improved Software Pricing on z990 Servers. Superior performance and
technology within the z990 has allowed IBM to provide improved software
prices for key IBM zSeries operating system and middleware software
products. The superior technology includes superscalar processing, native
"LPAR-only" mode of operation with the capability to support 30 LPARs,
support for multiple logical channel subsystems within a single system
footprint, and complete support of z/Architecture. While the announced z990
MSU values remain unchanged, IBM is pricing the designated software on z990
at lower MSU values than those previously announced. In order to implement
this IBM software pricing initiative on z990, IBM intends to lower the MSU
values generated by the z990 microcode by approximately 10 percent."
Yup. Because - despite the "superior technology" - the box was underperforming.

(It happens. Both HDS' Skyline and Amdahl's 5995M were hit by similar problems. G7 also had
performance hits with certain software because of the HSB structure. Business as usual. 10%
is nothing close to what you need to make z/OS competitive - it's a fudge.)
Post by Al Sherkow
MSUs have all the same traditional problems as MIPS, with one major
difference. IBM publishes MSUs
(http://ibm.com/zseries/library/swpriceinfo/hardware.html) and IBM never
published MIPS; they have been doing relative capacity for years with LSPR.
Bolleaux. You and I have been in the same room (breakout session at the Pallisades in 2003)
when IBM put MIPS charts up on the wall without a mention of either MSUs or LSPR.

See Figure 9.1 on page 31 of SG24-6863-00 - IBM eServer zSeries 990 Technical Introduction.

And Figure 7 on page 74 of the current LSPR manual - I wonder just what the 2084-301's "450
units" are.

And straight from an IBM "internal use only" foil that somehow got to a customer - it's about
z890 granularity. Just for fun, the exact same foil was shown to the GSE meeting in Leipzig
this spring several days before the z890 announcement by Dr Klaus Goebel - I was just outside
the room - in a plenary session without any kind of non-disclosure. So don't tell me IBM
"doesn't publish MIPS" - they publish almost nothing but:


z890 Granularity
MIPS / Capacity Setting
Uni's 2-ways 3-ways 4-ways
26/110 50/210 74/310 97/410
46/120 89/220 131/320 172/420
88/130 170/230 250/330 328/430
110/140 212/240 312/340 410/440
172/150 332/250 489/350 642/450
210/160 406/260 597/360 783/460
366/170 707/270 1040/370 1365/470
Post by Al Sherkow
It is consultants and IBM-Watchers (Gartner Group, Meta Group, Cheryl) that
publish MIP tables. IBM only wants you to use LSPR, but your management
probably wants a single number. MSUs are a single number for each machine
model. The LSPR provides information that matches our collective
experience, i.e. "not all workloads run with the exact same performance on
a given machine model. This is partly why the people who sell MIPS tables
usually provide a range of performance. I do not have Cheryl's or Gartner's
tables in front of me, though I've seen them in the past and they included
a range of performance. I believe they do not do their own testing, their
ranges are based on the LSPR, anecdotal information, and experiences from
their clients and customers.
My MIPS tables are free. They're worth less than what you pay for them. Asking money for
them would be a disgrace. The disclaimer on, e.g.,
http://www.isham-research.com/mips_z990.html should be taken at face value.
Post by Al Sherkow
The old-timers will remember that years ago
large organizations actually "benchmarked" machines. I don't think any
company does this anymore except for the vendors.
You think they do? I'd love to know EXACTLY to what extent LSPR is based on real
measurements. IBM doesn't tell us, and clearly states in LSPR that some results are
extrapolated/interpolated. My HP-41CV is very sceptical about some of the "measurements". I
still believe the original z990 LSPR was based on only three benchmarks - 1-way, 3-way and
16-way.
Post by Al Sherkow
Benchmarking is too
expensive. Even IBM does not test each model relative to each other model,
they compare a representative sample so customers can use the LSPR to
compare various machines and models. Gary King has an excellent
presentation on MIPS, MSUs and LSPR. His long experience in capacity
planning and working for IBM is nicely pulled together in this session.
One customer in the USA seems to have heard a funny story about MP factors and redesigns and
2008. Ain't anonymous email wonderful? Hotmail.com is a power for human advancement.

--
Phil Payne
http://www.isham-research.com

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to ***@bama.ua.edu with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
Phil Payne
2004-11-17 13:57:42 UTC
Permalink
Post by R.S.
Post by Ted MacNEIL
IBM publishes MSUs and IBM never
published MIPS; they have been doing relative capacity for years with LSPR.
IBM Canada always talked MIPS when I was at ScotiaBank. I still have a chart that they
gave me that had a 3090-200E at a relative capacity of 32 Units. I was amazed at that
'coincidence' (8-{]}

See Figure 7 on page 74 of the current LSPR manual - I wonder just what the 2084-301's "450
units" are.
Post by R.S.
Not only IBM Canada. I have a chart on the wall, printed by IBM (US I
guess), it's titled "Optimized for serious e-business", contains bars
and numbers - *in MIPS*.
The document number is G511-4172.
IBM UK produces a similar chart. I have several versions. And a cardboard slide rule with
MIPS on it. Also the pricing levels for emulation on Intel are at least informally defined in
MIPS.

IBM's "SP1 and S1 Price Guide" distributed to resellers mentions "MIPS" 64 times - including a
long discussion about replacing "under 40 MIPS" systems. The idea that IBM doesn't use or
publish MIPS is frankly ludicrous - they rarely use anything else.

--
Phil Payne
http://www.isham-research.com

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to ***@bama.ua.edu with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
Phil Payne
2004-11-19 15:21:32 UTC
Permalink
Post by Al Sherkow
IBM really changed this discussion when the z990 had one Announced
MSU size, and a smaller MSU number for Software Pricing.
I wonder why they didn't just drop the price per MSU by 10%.
Apart from not promoting new hardware, that would be a very different thing.

Everyone's mileage will inevitably vary (very much) but imagine z/OS running on a Sysplex of
two 2084-312s. Initially announced at 691 MSUs each, 1382 for my hypothetical Parallel
Sysplex at full load.

1382 is just inside VWLC Level 6.

Post repositioning, the 2084-312 becomes a 620 MSUs machine and my Sysplex becomes 1240 MSUs.

The reduction in charges seen by the customer is thus the equivalent of 67 MSUs charged at
Level 6 prices and 75 MSUs at Level 5 - because of the degressive nature of the VWLC price
bands, nothing like 10% off the software bill. It's not correcting software
price/performance - you'd need 90% to do that anyway - as with NALC and z/OS.e. It's
compensating for z990 underperformance.

Note this has a rather perverse side effect - imagine we were talking about a DB2 system and
the customer wanted to add a new application. The new application would cause MSUs to be
added at Level 5 prices rather than Level 6 - so each MSU for the new application would be 30%
more expensive than before the repositioning. The total bill, of course, is still lower.

There is no long term effect on zSeries software price/performance - organic growth will
swallow this up very quickly.

--
Phil Payne
http://www.isham-research.com

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to ***@bama.ua.edu with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html

Loading...