I've seen a number of people speculating in this thread about why z/OS's
market share hasn't increased, covering everything from technical
features, sales, hyperscalers, etc. I think that these are all good
points, but they are missing the real underlying factor as to why
adoption hasn't increased. This
(https://web.archive.org/web/20070605070840/https://money.cnn.com/magazines/fortune/fortune500_archive/full/1995/)
is a list of the fortune 500 in 1995. Almost all of these companies were
mainframe users then, and still are today. What changed is that new
companies stopped adopting mainframes.
While previously, any company of decent size that wanted reliability and
performance over a certain threshold would have to hit up their local
IBM sales representative, this changed with the PC revolution around
1995ish, when Linux or Windows NT combined with high speed networking
made it possible to achieve decent reliability and
decent performance for a fraction of the cost. Critically, they also
allowed you to start small! This
(https://commons.wikimedia.org/wiki/File:Google%E2%80%99s_First_Production_Server.jpg)
was google's first server rack! The discussion about weather or not
google would run better on a mainframe is pointless. Google's first
servers were a bunch of home brew computers attached to a surplus rack.
In no universe would they ever have been able to afford a mainframe, and
the IBM sales rep would have likely laughed them out of the room.
This meant, that as google grew, it grew around the clustered Linux
platform that it started with. That became it's legacy. In the same way
that mainframe -> Linux migrations are almost impossible for a major
bank, a Linux - mainframe migration would be just as impossible for
Google. There's too much tech debt, and too many processes that have
adapted to the old system. The technical details are almost irrelevant.
Tomorrow, an angel could visit IBM headquarters and deliver a set of
golden CPU dies and OS source tapes, to create a mainframe system
with performance 100 times anything else currently on the market and it
still wouldn't matter, because IBM would price it at a start price of a
million dollars, and the only customers would be those doing lateral
upgrades.
I've said it in SHARE talks, I've said it in person, and I've probably
said it on this mailing list, but the only way that IBM is ever going to
grow the mainframe platform, solve the skills gap, or even just stop the
slow gradual erosion of customers and installed sites is if they create
a machine cheap enough, and with license costs low enough, that /new
startups can afford it./ IBM must bring in customers who don't yet have
legacy, and that means startups and new customers. The most effective
way to sell mainframes right now would be for IBM to create a box that
costs 10K, takes up 4RU, and can run the equivalent of a few dozen
decent sized EC2 instances. This would be useless for every existing
customer base, but it would be affordable to startups, who are don't
want to get annihilated by AWS/GCP/Azure costs, and haven't yet built
out their infrastructure.
Until we have that, discussions of feature matrices, reliability,
instruction speeds, or the sales force is like rearranging deck chairs
on the titanic.
Just my two cents on the topic,
Enzo Damato
On 10/25/25 10:16 AM, Jon Perryman wrote:
On Fri, 24 Oct 2025 16:23:05 +0000, Dick Williams<[email protected]>
wrote:
Why would IBM want to sell a mainframe (zOS) to Google?
Do they process important transactions?
This mindset has doomed z/OS. These are the wrong questions. IBM (International
Business Machines) wants to sell to all businesses.
Not all businesses are about important transactions. Instead, ask why isn't
Google replacing their 5M Linux servers with 250,000 z/OS servers? What
business problems does IBM solve that are important to Google?
A single Google search moves hundreds of strings. The z MVST (move string) is a
single instruction residing in cache. On the other hand, the STRCPY() function
for other architectures is several instructions that reside in storage and must
be moved to cache as needed.
A Google search is heavy disk access. MVS has been running as a NAS since the
1960s.
Google developed GO 15 years ago but missed important and simple features (e.g.
recently discussed PRINTF compile time type validation).
You get what you pay for and Google pays $0 for Linux, GCC and others.
Linux is looking to eliminate big endian support despite it being of little
impact to Linux. Most endianess is handled by hardware instructions and is very
little impact to programs.https://www.phoronix.com/news/Torvalds-No-RISC-V-BE
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email [email protected] with the message: INFO IBM-MAIN
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN
