Index Home About Blog
From: rpeglar@csinc.UUCP (Rob Peglar)
Newsgroups: comp.sys.super,comp.sys.cdc,comp.unix.cray,comp.arch
Subject: The ETA Saga
Message-ID: <198@csinc.UUCP>
Date: 17 Apr 90 21:07:37 GMT
Followup-To: poster
Organization: Control Systems, Inc., St. Paul MN
Lines: 678

Disclaimer:  The following text is the author's opinion only.  It neither
construes nor represents any policy of Control Systems, Incorporated or
Comtrol Corporation.  

This is somewhat long.  If you don't care to read this, please hit "n".
I have cross-posted to a number of newsgroups, as there are people who
may be interested that do not subscribe to comp.sys.super.  Note to
comp.arch'ers - this is not traditional comp.arch material.  Enough.

			The ETA Saga


		How to (Mis-)Manage a Company
		According to Control Data Corp.

"We have met the enemy, and he is us."  -  Walt Kelly, from "Pogo"

			Part One

Chapter I - The Last Day : April 17, 1989

Let us begin, oddly enough, at the end.  

I awoke early, around 5:30 AM.  It was a rather cold, early spring day, with
a slight hint of frost.  This Monday morning, it seemed to me at that early
hour, was full of promise and excitement.  The night before, I was startled
out of a light sleep at 10:45 PM by the phone.  Groping for the receiver,
I picked it up and listened while a voice encouraged - no, insisted - that
I attend a meeting of all ETA managers the next morning, at 7:45 AM in the
CDC headquarters complex.  The voice then wished me a good night.  Feeling
that a major change - for the better - was about to take place, I went back
to sleep feeling very good indeed.

I arrived at the ETA headquarters around 6:30 AM.  The parking lot was almost
empty.  Still feeling excited, I strode to the West entrance only to find
the security guard standing outside, a very unusual occurance.  He stopped
me and asked to see my badge.  I showed it, and he said "I'm sorry, you 
can't go in."  He had a very short list of names on a clipboard, and refused
my request to look at it.  The doors were locked.  Glancing at the list, I
noticed the very top of the four-level ETA management structure appearing
on it with very few others.  I then quickly turned away and headed back to
my car.  The only thing to do now, it seemed, was to proceed to CDC a bit

I was the first person who tried to enter the conference room where the
meeting was to take place.  Again, I was refused entrance, and was told
that at 7:30, the room would be opened.  I met one of my fellow managers
and we chatted, albeit nervously.  A few at a time, the other managers,
about 100 people in all, arrived and seemed uniformly nervous.  I still
was convinced that something good was happening, and the locked doors
and such were CDC's strange way (CDC had strange ways about almost
everything in terms of the way they handled people) of getting us to go
to this meeting.  By 7:45, we were all assembled and sitting down.  Mr.
Price strode into the room with a paper in his hand.

In the next ten minutes, we were informed that "we were no longer managers
of ETA Systems" and that "we were terminated effective June 17th".  Mr.
Price took no questions.  He quickly left the room.  The ETA VP of
Human Resources (aka personnel) then mumbled things about packets and
unemployment and procedures.  The entire meeting lasted exactly 12 minutes.

The reactions from the group ranged from "I told you so" to disbelief to
anger to tears.  I suspect there were 100 different reactions amongst the
100 people.  I was stunned beyond comprehension.  I had expected the
meeting to announce the sale of ETA to one of many interested parties,
many of whom had been in the ETA building in the two weeks prior discussing
tactics with the executives.  Instead, I tried to rationalize the decision
based on all normal criteria, and I could not.  CDC had performed the
ultimate stupidity, the ultimate lobotomy of their own computer business

We left, heading for the World Theatre in downtown St. Paul.  The entire
company (sans management) had been called early in the morning and told
to assemble at the World at 8:30 AM.  Arriving at the theatre, we trooped
in without saying a word.  All 800 (or so) employees were there, and
the noise level was high.  People seemed as excited as I had been earlier
in the morning.  For us, however, it was like attending your own funeral.
We were not allowed to say anything to the employees.  So, we sat high
in a balcony and waited.  Almost as if a clone of Mr. Price, Mr. Perlman
appeared on the large stage and the crowd because hushed.  

Perlman read much the same text as Price had one hour before.  Ironically,
when he mentioned that "you are considered locked out and should not report
for work" there was a fair amount of cheering.  Perlman also took no
questions, leaving quickly as if a phantom was chasing him off the stage.
The VP then rattled off the same meaningless words about displacement and
procedures.  No one listened;  everyone was in their own private hell.

Thus, at about 9:00 AM on that chilly April day, the world of supercomputing,
the world of American technology, the world of smart business practice,
the world of 800 people - changed irreversibly.  What follows is one
opinion on why.
Chapter II - The Beginning

ETA began operations in September 1983.  Born from a skunk works-brainstorm
group of six CDC people, ETA was led by Lloyd Thorndyke, a long-time CDC
executive and head of the CDC Advanced Design Laboratory.  Neil Lincoln,
the chief engineer on the Cyber 205 project, was the technical leader at
ETA.  The original goal of ETA was straightforward - to design, develop,
manufacture, and ship the next generation supercomputer, relative to the
Cyber 20x series, in the least possible amount of time - hopefully, three

To reach that challenging goal, ETA's Gang of Six convinced CDC brass
(at that time William Norris, Bob Price, and Henry White) that the only
way was to "spin off" a separate corporation.  Spinning off would allow
three things to happen - three things, as it turned out, that never
occured.  First, ETA would function as an independent corporation.  Its
stock was to be sold publically.  CDC would hold 40%, a minority share,
of the stock.  Second, ETA would try to access government funding - 
research/development dollars from various departments, e.g. DoD, DoE,
and agencies like NSF, etc. - to help it in the formative, capital
intensive years.  Lastly, ETA would be run by its own board of directors,
and have little if any CDC "interference" from either an executive
management or, more importantly, technical point of view.

All of these tactics had been tested and validated a decade before
in much the same scenario.  In 1972, however, the players were Cray,
Davis, and others from CDC Arden Hills.  There were great similarities
between the ETA endeavor and the formation of Cray Research - but the
timetables were much stricter for ETA.  Six years' worth of R&D had
to be accomplished in three years' time - it had taken Cray until
1978 to ship the first Cray-1.  The expectation (indeed, the 
announcement from CDC) was that it would take ETA three years to ship
the first ETA-10.  Actually, the first announcement by CDC about ETA
was contained in a speech given by Norris at Los Alamos in August 1983. 

The first major business failure and the beginning of the second major 
business failure by CDC occurred in the first three years.  ETA, acting 
largely on its own (by design), could not interest the government in any 
funding activity.  Also, by then, CDC had begun to change its corporate
mind about taking ETA public.  Thus, CDC had shifted the entire risk onto
itself.  However, CDC never publically acknowledged this;  the CDC
corporate stance was to give ETA most of the responsibility and little of
the authority.

In the first three years, ETA actually did manage to design, develop,
manufacture, and ship an ETA-10 supercomputer.  However, the strain of
sole funding from CDC on this extraordinary event would prove to be
costly.  There was a large media event on December 31st, 1986, when
the trucks rolled from ETA to Florida State, containing ETA-10 Model E
S/N 1.

On the technology side - hardware - this was truly a remarkable event.
Even though it needed constant care, the machine actually worked.
However, from the software side, this was truly a non-event.  There
was no operating system software, as commonly defined.  Florida State
(FSU) took delivery of a machine that only ran in monitor mode, in
single-tasking, single-user (literally, one person at a console - a
throwback to the 1950's era of "open shop" computing), loading raw
programs one at a time from an external source (an Apollo workstation),
and re-booting after each program completed.  Programs and the "operating
system" - at that time, a collection of hardware support routines -
were compiled and linked in a cross environment using both Apollos
and Cyber 205s.  This was the state of the EOS art after three years.

This was the first major technological failure - the insistance of
(trying to) develop an entirely new operating system, from scratch.
EOS was a failure due to many reasons, but CDC played a major part in
the failure.  That part was the lack of market research concerning Unix
and its role in supercomputing.  Cray Research (CRI) had already shipped
two releases of Unicos, its AT&T Unix System V Release 2 derivative work.
Although ETA Software R&D was the instigator behind EOS, CDC was the
entity that took the bait.  This led to many disasters in the coming
years.  CDC was convinced that EOS was the only path to allow their
Cyber 205 installed base to upgrade to ETA equipment.  Even though the
entire world of scientific computing was moving toward Unix, CDC stuck
to the old proprietary formula.  To ETA's credit, a push - albeit much too
small, in terms of actual resources devoted - toward a "Unix environment"
under EOS was begun.  Also, to bring the first three years to a close,
an "underground" research project, dubbed "Baron Von Cygnet" was begun
to study the feasibility of Unix System V native on the ETA-10.  It was
this skunk-works-like research which paved the way into the second,
and last, three years.  This research project was instigated by ETA itself,
urged on by a leading voice in the supercomputing community, NASA/Ames.
Chapter III - The Middle and the End

The middle period at ETA marked major upheavals in the top two layers
of ETA management, as well as the top three layers at CDC.  At CDC,
Norris retired as CEO and later Chairman.  Price became CEO (a post
which he just recently retired from) and White retired.  CDC brought
in Thomas Roberts to head Computer Products, and Gil Williams to head
Cyber and ETA products.  Both Williams and Thorndyke reported to Roberts.
Gradually, Williams and Roberts worked together to effectively reduce
ETA's ability to manage itself.  In November 1987, the most important
ETA Board of Directors meeting to date took place.  During that meeting,
Roberts (the lead gun from the CDC contigent on the ETA board - which
dominated in voting power) announced that (paraphrasing from Thorndyke)
"it's our (CDC's) company, we'll run it the way we want to."  At that
point, Thorndyke was told he was being replaced by Carl Ledbetter, who
had stints at Prime and later IBM as head of the vector facility (VF)
project.  Roberts had made it abundantly clear that ETA was no longer
in charge of its own destiny.  This was the third, and final major
business mistake by CDC - the death blow, as it turned out.  This
strangulation of ETA was a premeditated move on the part of Roberts
and CDC.  

Ledbetter's presence at ETA created a considerable rift, at first.  
Thorndyke was extremely upset by being replaced and herded off into
a corner.  Now, Ledbetter was a very stylish and reputable leader -
a visionary, according to some.  His many talents were obvious.  He
was a dynamic speaker, and could turn into the world's greatest salesman
at a moment's notice.  Yet, he was fluent in most all technical aspects
of supercomputing, save software.  Within ETA's walls, his power was
absolute.  However, it was easy to see that his cards were actually
being dealt by CDC.

During 1988, Ledbetter went about assembling his own team of executives
to run ETA.  He brought in two former IBM'ers, Jones and Moschner, to
run sales/marketing and operations, respectively.  Neither man had
any direct experience in supercomputing, a fact which led to further
division at ETA, espcially amongst those loyal to Thorndyke.  It was
1988 which proved to be the pivotal year for ETA - the first time that
it became obvious that ETA could be successful.  There was, in fact,
much light at the end of the tunnel.  The successive introductions of
the ETA-10P - the air-cooled supercomputer - and ETA System V, which
had grown out of the skunk works into a workable product - turned the
world's head toward ETA.  At Usenix Summer 1988, ETA brought a live
ETA-10P to the show floor, running ETA System V, and it worked.  Many
were absolutely astounded.  No one had ever shown a live supercomputer
on a show floor before.  It was a marketing and technical coup, brought
about by a team of marketeers including Meckes and Gardner, and the
technical prowess of the ETA System V group led by Torkildson.

The company actually became "profitable", in a sense, during 1988.
Measured as an R&D house only - i.e. revenue from products sold minus
R&D costs - ETA was plus ten million dollars, approximately, in 1988.
By years' end, 27 systems - 20 air, seven liquid - were installed
around the world.  The liquid systems were installed as follows:

	* Florida State University
	* John Von Neumann Center (JVNC) for the Consortium for
	  Scientific Computing (CSC) - two systems
	* Tokyo Institute of Technology
	* Deutscher Wetterdienst (DWD), the German Weather Service
	* UK Meteorlogical Service (UKMET)
	* Minnesota Supercomputer Center (MSC)

The air-cooled systems were scattered around the world, with systems
in the USA, Canada, Europe, Japan, and Australia.  Almost all of the air-
cooled systems ran ETA System V;  the liquid systems were split between
EOS and System V.  However, every site that was running EOS - save DWD -
had a plan in place to convert to System V during 1989 and/or 1990.
Florida State led the way, converting their model E over a weekend to
System V, with little difficulty, in late 1988.

Given all this activity, which had never been accomplished before in
the supercomputing industry (remember, it took CRI six years just to
ship the first Cray-1), 1989 seemed to be the year for ETA to actually
break even in terms of a true business.  Spirits were very high amongst
most ETA people.  ETA was actually mentioned favorably in the CDC fourth
quarter report of 1988, in remarks from Price.  

However, there was a brewing kettle of trouble for ETA at CDC headquarters.
Roberts and Williams were stewing over ETA's successes.  Relative to the
dismal year - and even more dismal future - for the Cyber mainframe
business, ETA was a shining star.  This reality irked Roberts and Williams
to no end.  Roberts, in particular, was annoyed by the favorable press
coverage given Ledbetter and ETA in general.  Roberts went so far as to
have ETA's major press/PR consultant, Gil Goetz, fired because of the attention
paid to ETA and Ledbetter in particular.  The more momentum ETA acquired,
the more Roberts and Williams stewed.  Roberts' placing of Ledbetter
as the helm of ETA had actually "backfired" in a sense - now the son
(ETA) was actually out-performing the father (CDC) in the computer business.

This was apparent throughout CDC's sales force, who were charged with
selling both CDC and ETA product lines.  The majority of the sales force
had reached a dead end on the CDC line; thus, they focused on selling
ETA products.  This situation, however, was a "catch-22" due to the over-
whelming lack of knowledge of ETA and supercomputing in general in the
CDC sales force, leading to even more pressure on ETA R&D people to
assist in the sales process.  

During late 1988, CDC decided to try to leverage ETA's success by pulling
the company back into itself.  ETA stock was bought back by CDC and ETA
was made a wholly-owned subsidiary of CDC.   Roberts was cast aside by
Price - the official statement read Roberts would "pursue other interests" -
and Perlman was put in place as COO, Ledbetter reporting to Perlman.  (Price
later retired as CEO but was still chairman of CDC's board).  Williams remained
as head of Cyber products, also reporting to Perlman.  However, the seeds
of corporate discontent sown by Roberts were soon to bear fruit, with
Perlman as the gardener.

Early 1989 was a dichotomy of actions by CDC and ETA.  ETA, on one hand,
continued to ship large (by supercomputer standards) volumes of systems,
almost all air-cooled.  The ETA-10 P and Q models were gaining great
favor in the industry, overwhelming Cray and Convex in price/performance
by a wide margin in many instances.  CDC, on the other hand, was obsessed
by its own failures in the mainframe business;  for example, approximately
75 million dollars was sunk into the R&D efforts behind the Cyber 990
series, and only 12 systems were sold in 1988.  Compared to the 50 million
dollar R&D effort at ETA, which resulted in 24 systems sold in 1988, CDC
knew it had to do something.

Predictably, in classic CDC form, Perlman had decided the short-term bottom
line was the only criteria to be considered when measuring success.  ETA's
bottom line was still very negative - cost of sales alone in 1988 was
80 million dollars - and Perlman could not tolerate the situation.  Even
though all the factors were positive for ETA, CDC had already decided
to give up on ETA.  A buyer was to be found, and very soon.  ETA was
to be sold.

In February and March of 1989, CDC was consumed with this process.  
Ledbetter, Moschner, and the other executives at ETA were also consumed,
spending many hours in closed-door sessions with potential buyers of
ETA.  It was known that Unisys was extremely interested;  other companies
rumored to have interest were Ford, Chrysler, and TRW.  (Ford and TRW
actually had procured ETA air-cooled systems for themselves).  However,
the was one rather large catch - the asking price, as set by CDC, for
ETA was rumored to be as high as 500 million dollars.  The current
value of ETA was considerably below that, and many thought the future
value (5 years, say) was only around 200 million dollars.  

This discrepancy led to stalement and refusals by each and every
potential suitor of ETA.  CDC held firm on its asking price, trying
to milk every last dime out of the sale of its own child.  There was
hope, even up until April 12, 1989 - the Wednesday before the Last
Day - that ETA was to be sold.  However, on that day, all hope was lost
as the last potential buyer walked away shaking their heads at CDC's
ridiculous asking price.

Ironically, on April 14, 1989 - the last day of operations for ETA,
two major developments occurred simultaneously withing ETA R&D.  First,
R&D had decided amongst themselves, certainly encouraged by Ledbetter,
to abandon its pursuit of EOS and what it stood for, in favor of throwing
all available resources onto ETA System V.  Second, ETA System V Release
1.1 was signed off into production, a step which would allow the
final formal acceptance on many installed (but not formally accepted)
systems, leading to a hefty increase in revenue for 1989.  These two
events would have led to yet another era within ETA - an era, as it turned
out, which lasted exactly one weekend.
			Part Two
			...and Analysis

Chapter I - The Business of Technology

ETA Systems, Inc. failed as a business due to three major business, or
management-related, reasons, and one major marketing reason.  As
noted in Part One, the three major business reasons are:

	*  CDC's failure to carry out its plan to take ETA public
	*  CDC's failure to secure government funding for ETA
	*  CDC's failure to establish and manage ETA independently

The major marketing reason is:
	*  ETA's/CDC's failure/refusal to recognize the market demand for 
	   Unix operating system in time

There is absolutely no comprehensible reason for the management-related
failures to occur.  All three points formed a large part of the original
basis for ETA's conception and existence.  Any one of the three failures
occurring would have been a severe blow to ETA as a company;  all three
occurring was fatal.

The first two failures concern capital and its raising.  ETA as a public
entity would have allowed the market, as a whole, to decide the
viability of ETA's product.  ETA funded by government entities would
have greatly contributed to a new philosophy of managing industries
important to the nation.  There are many examples of this in various
Japanese industries.  In either case, such funding would have relieved
the cash pressure on CDC, already burdened heavily by its 250 million
dollar debt offering of 1986.  

The third failure was an error of commission.  CDC became very aware
that ETA, by 1988, had the ability to be a large commercial success.
After the ETA Board of Directors meeting in November 1987, however,
Tom Roberts assumed total control over ETA as an entity.  From that
point on, there was no chance that ETA would ever be managed independently
of CDC.  Just as the product was reaching the point of viability in
the marketplace, CDC decided to enshroud it in a heavy veil of corporate
red tape and overly bureaucratic procedures.  

CDC had failed the lessons taught by Cray in the early 1970's.  In the
capital- and intelligence-intensive world of supercomputer research and
development, the only possible postive solution is to form a small
company, secure public funding, and leave the engineers alone for a
good long time.  In Cray's case, it was six years until the first Cray-1
installation, and approximately nine years before Cray first earned
recognizable profits.  

A root cause of CDC's insistence that ETA mature (over-)rapidly and
earn profit for the mother company is the adjunct failure of the CDC
Cyber product line in the late 1980's to generate sufficient profit.
Put another way, ETA was looked at - especially by the sales force - as
a "savior" of the CDC computer business.  Cyber sales were down and
heading towards non-existence.  Even Gil Williams himself, during a
company-wide (CDC) Computer Systems meeting, at which all CDC and
ETA managers attended, spoke about the computer business as two
growing parts - workstations and supercomputers - and one shrinking
part - mainframes.  Both workstations and supercomputers were to see
double-digit growth in 1989; mainframes were to stay flat and actually
shrink slightly.  

The marketing error of commission was a joint blunder on both CDC's
and ETA's part.  During the first two years of ETA's existence, CDC
committed a horrible marketing mistake, and that mistake was then
followed through by two horrible ETA decisions concerning software

In 1984, CDC had decided that the operating system for the ETA-10 was
to be upwardly compatible, at the level of user source code, to the
existing Cyber 205 operating system, VSOS.  This compatibility included
many (but not all) commands, utilities, library calls, and special
CDC-isms in the compilers, linker, and I/O library.  This decision by
CDC was a huge mistake, one that had no basis in rational thought, given
the supercomputer marketplace.  CDC need only look over the river (the
Minnesota, that is) to its offspring, Cray Research, to find out what
changes were occurring in the supercomputer operating system world.

Cray in 1984 was finally beginning development on a version of Unix
System V Release 2 (owned by AT&T) for its existing product lines.
After two false starts by Cray, in 1982 and again in 1983, to begin
Unix development, the third attempt in 1984 went through Cray management
and was finally blessed.  Compare this to CDC in 1984, who decided
to quash all thoughts of Unix on ETA machines, and (yet again) re-
invent the wheel, starting literally from scratch, with its VSOS-ish
attempt at (yet another) propreitary operating system, called EOS.

Given this CDC imperative, ETA Software, at that time led by Dale
Hiveley and Jerry Campbell (as director and chief designer, respectively)
came up with a powerful and much too complex operating system.  Trying
to endow EOS with every feature and function known to supercomputing,
ETA wasted month after month in the design phase for EOS.  It actually
took nine months, from late 1983 into mid-1984, to decide what language
to write this monstrosity called EOS in - and the decision, predictibly,
was Cybil, a proprietary variant of Pascal, developed by CDC for its
Cyber product line internal software development.  This was one of
many predictibly horrible decisions made by ETA in the first three years
of software effort.  Instead of choosing C, the language of choice of
software developers, even in the mid 1980's, ETA chose Cybil and CDC
blessed it (of course).  

Decisions like Cybil instead of C led to massive delays in developing
EOS in the early days.  ETA did have a classic development problem -
no native hardware - to contend with.  Faced with this daunting road-
block, ETA chose the path of most resistance - use the Cyber 205 as
a software development engine.  ETA also chose to use an Apollo work-
station environment, under Aegis, to hopefully make the horrendous
job of EOS development easier.  Predictably, ETA chose to use CDC
propreitary environments to base its development.  This choice was
another factor in the all-too-slow development of EOS in the first
three years.  The first sign of these terrible choices appeared, in spades,
at Florida State University during the first months of the operation
of the ETA-10 at an actual customer site.  EOS was non-existent at
FSU.  It took wave after wave of EOS developers, at massive cost to
ETA, to care for the monitor-mode, single-task, boot-after-execution
morass known as EOS.

The first true release of EOS occurred on February 12, 1988.  This
version supported up to two users per CPU, each with one batch job
or interactive session.  However, in the field, sites such as the
John Von Neumann Center (JvNC) quickly realized that even the extremely
limited usage (e.g. two batch jobs at once) was a strain on the
native EOS capability, and scaled back to one job at a time.  Thus,
after four and one-half years of software "development", the best
EOS could realistically offer was single-user, single-tasking support.
ETA-10 EOS users quickly dubbed their systems as "personal supercomputers",
or more quaintly "EOS - the OS that turns a supercomputer into a PC".  

By this time, Cray had shipped multiple releases of Unicos.  Although
rough around the edges, Unicos had won great favor amongst the super-
computing community for its simplicity and ability to make porting
of applications easier (than its own proprietary OS, COS, for instance).
Many ETA prospects, and even existing customers, wondered out loud about
ETA's plans for Unix.  Predictibily, ETA tried to appease them by
speaking about a "Unix environment on EOS".  This was pure rhetoric
until mid-1988.  EOS development failed to allocate enough resources
onto the Unix environment project, rationalizing this with its own
problems on native EOS development and support.  It was of this user
community frustration that native Unix on ETA-10 systems was born.

In late 1986, Nasa/Ames and its NAS facility had spoken, somewhat under
the table, to ETA about developing a native Unix capability.  This
negotiation led to a "skunk works" research project, known as "Baron
Von Cygnet", headed by Eric Rowe, who had broken away from the EOS
design group.  This research quickly led to an effort by ETA to
establish a proof of concept system.  Realizing that ETA had no one
in house to develop such a system, ETA turned to outside help, in the
form of Lachman Associates, Inc. (LAI), a Chicago-based Unix development
contracting firm, to begin actual development.  Also, ETA made the
decision to begin its own Unix development group.  The leader was
chosen from the outside (of ETA and CDC - the first sign of a rational
decision) and the project quickly bloomed into fruition.  In only
six months after the development go-ahead, the Unix group - less than
five ETA employees coupled with approximately ten LAI people - had
native Unix up and running native on the ETA-10 in single-user mode.
Compared with the 4 1/2 year effort to accomplish the equivalent in
EOS, the Unix effort had three major factors in its favor:  it was
relatively simple (compared to EOS); it was inexpensive in terms of
R&D cost; and it was what the market wanted.

Chapter II - The Technology of Business

From the inside view, ETA operations could be characterized as covering
a very broad spectrum of efficiency.  Bluntly, the various departments
at ETA were anywhere from wonderfully innovative and visionary to
blindingly archaic and bureaucratic.  The latter characteristics, predictably,
were remnants of the CDC history and influence upon ETA.

The sales group is a case in point.  In the beginning, the objective was
to have an independent sales force, only charged with selling ETA products
and services.  In the middle of ETA's six-year life, CDC changed its mind
about the sales force, and insisted that ETA use only CDC sales people.
This transistion wasted months of time and hundreds of thousands of dollars.
At one point, ETA had a seven-member sales force.  Unfortunately, this
was during the middle years (1985-1987) when ETA had very little product
to demonstrate and sell.  It was 1988 before the sales force was consumed
by CDC entirely.

CDC sales people had very little knowledge, generally speaking, about
supercomputing in general and ETA products in particular.  CDC paid
lip service to sales training.  However, the sales force had little or
no enthusiasm for selling Cyber products - a phenomenom which encompassed
the CDC sales force worldwide - and therefore tried any and all tactics
to sell ETA products.  The natural result of this "selling wine before
its time" were gross misrepresentations of the system to the customer(s),
wildly optimistic performance characterizations, shipments dates which
were pulled from thin air, etc.etc.etc.  Many customers were left with
massive unmet expectations.  The classic case in point on this unfortunate
situation was the Tokyo Institute of Technology.

The Tokyo order was for a hardware configuration which had never been
attempted before, running a non-existent OS, using non-existent tools
(e.g. FORTRAN compiler), to run non-existent applications.  The order
actually came down in fall 1987, for an ETA-10E8, with native Unix, to
run a series of applications including NASTRAN.  Notice the date -
native Unix had barely been booted into single-user state by then. The
compiler(s) and applications were just a dream.  The machine itself
(8 processors) was only a drawing-board exercise in lines and boxes.
Yet, very precise reliability, availability, performance, and 
maintainability numbers were quoted in detail in the proposal to the
Institute.  How could this be?  Could the ultimate sale of vaporware
actually come to pass?

It could and it did, due to immense pressure from CDC on ETA to book
this order.  There were many government and commercial entities involved
in this order, from ETA and CDC to the Department of Commerce (US) to
MITI (Japan).  All parties involved wanted a supercomputer - made in
the US, by someone other than Cray - to be sold and delivered to the
Institute.  There were also many influences on the Japanese side, like
the school rivarly between the Institute and the University of Tokyo -
who had Cray equipment up and running.  All of these "business" - i.e.
non-technical issues - forced ETA's hand.  

Of course, the order should have never been accepted.  The salesman
was out for his commissions.  CDC Japan wanted very much to have their
first computer system - their first, the world's largest machine ! - 
in their territory.  CDC Japan was an unbelivable combination of lack
of technical knowledge and "insider" business practices.  Note, CDC
Japan no longer exists today, thankfully.  Even at the last day of
ETA, CDC Japan was withholding one-third - 9 billion yen - of the
payment on the Institute system due to a "gentlemen's agreement" between
the president of CDC Japan and the professors in charge at the
Institute.  The system, in its entirety, should never have been accepted
in the first place (May 1988) by the Institute.  The immense pressures
from outside agencies, however, led to its eventual acceptance, with
two-thirds payment.  The entire affair was horribly botched, due
to the singular factor of all parties involved - that is, all parties
with authority but no responsibility - to get the system installed
and accepted.  The parties with responsibility, of course, became
innocent pawns in this affair of international commerce and intrigue.

ETA Marketing was also quite interesting, although in a much different
way than sales.  The marketing group actually had quite a few very
good people;  their hands were tied very tightly, however, by CDC
marketing types who had little knowledge of what it takes to market
a supercomputer today (as opposed to ten years ago).  Budgets, strategy,
and press coverage were three examples of CDC interference, for the
worse, on ETA marketing.  The budgets allotted ETA marketing were
pitifully low.  Remember, CDC called the tune on each and every ETA
department budget - marketing, software, manufacturing, technology,
and engineering.

The technology and engineering groups were similar to marketing in
that they had quite a few good people.  However, the engineering group
as a whole had resources fairly skewed in terms of sub-groups.  Most
noticeable was the lack of resources dedicated to the I/O subsystems
of the ETA-10.  The engineers who had detailed knowledge about I/O
subsystems could be counted on one hand, using one finger.  There were,
however, a plethora of CPU experts.  The CPU was rarely a problem
either in the field or the lab;  the I/O subsystems were always producing
subtle and sometimes disastrous errors, especially in the interface
between the data pipe controller (aka the "light pipe") and the shared
memory.  One of the most common failures of an ETA-10 was that a
light pipe had to be physically replaced.  

The software group was a quite interesting group of people.  Its structure
provided many lessons in how not to organize a coherent R&D effort.
The EOS group, at its peak, was nearly 100 people strong.  Quite a few
of the group had more than passing experience with the Cyber 205 and
VSOS.  The Unix group, at its peak, was about 35 people.  Very few, if
any, had direct experience with the 205 and/or VSOS.  This actually
turned out to be a great benefit.  

There was constant tension between the two groups (EOS and Unix).  
Physically, machine resources were very tight;  another side effect of
limited and constrained R&D budgets handed down from CDC.  The two
groups often needed each others' machine resources.  It was quite
easy to switch from EOS to Unix, but the reverse could be quite difficult
in terms of effort and time involved.  Also, there was a constant clash
of personalities and philosophies between the groups.  EOS always took
pride in their work, and was often offended by the Unix group's less-
than-benevolent avoidance/ignorance of EOS.  Put another way, the
Unix people could care less about EOS.  Some of this was quite
justified - EOS was clearly a market failure, and Unix was clearly
a market success (the count of systems on The Last Day was Unix 23,
EOS 11).  On the other hand, EOS people were cognizant of this and did
care about the effect Unix had on ETA in general, but were put off
about Unix as an OS, quoting many of the same ill-conceived notions quoted
by many Unix-bashers past and present.  This philosophic battle actually
did materialize itself in a fistfight one night at a bar close to
ETA.  That was the only time, fortunately, that a member of the EOS
group and a member of the Unix group had such an encounter.  

In conclusion, the demise of ETA was a direct result of CDC management
intervention when inappropriate.  Almost universally, CDC management
decisions, when applied to ETA, were incorrect.  Many errors of both
commission - for instance, the virtual takeover in November 1987 - and
omission - for instance, the lack of market recognition for Unix on
supercomputers - occurred directly from CDC mismanagement and over-
zealous dogma.  The truly sad fact about the demise, indeed, murder of
ETA as a company is that its failure reflects not the failure of
technology, but the failure of people, more specifically top managers.
Many people in the computer industry assume that most computer company
failures must be a result of poor product, design, or manufacture of
some kind.  Not so - computer companies fail because of poor management
and erroneous, ill-timed decisions - or the lack of coherent, timely

The failure to take ETA public, the lack of government funding, and
the over-protective management applied by CDC were death blows to
ETA.  Issues of product quality and support pale before such important
business issues.  Ledbetter put it quite nicely in an interview with
a local Twin Cities journalist:  "CDC took a field of corn, days before
harvest, and plowed it under".  The agriculture analogy is quite 
appropriate in many ways, the least of which is the fact that CDC and
ETA were Minnesota-based companies.  Another analogy which has been
attributed to an ex-ETA person is this: CDC, an alcoholic father, who
seemingly means quite well, attempts to give his child a hug, but in his
alcoholic stupor, ends up smothering the child;  death by suffocation.

The lessons for the American computer industry are many.  It is the
fervent hope of many that the lessons are learned by those looking
toward the future.  

Rob Peglar	Control Systems, Inc.	2675 Patton Rd., St. Paul MN 55113
...uunet!csinc!rpeglar		612-631-7800

The posting above does not necessarily represent the policies of my employer.


Index Home About Blog