🚀 go-pugleaf

From: ihnp4!utzoo!henry@ut-sally.UUCP

(John, you might as well remove me from the mailing list, partly because
I get it via news but partly because we're no longer looking seriously
at the Sequent machines.  The latter is the reason I'm sending this to
the general mailing-list address; I thought our reasons might be of
interest.)

Our new-machine purchase is turning into a drawn-out drama because our
upper management hasn't yet produced the cash.  What's dramatic about it
is that as a result, we've had time to change our minds a couple of times
about what we want to buy.  If we had been handed the money last July or
so, we'd probably have bought a Balance 8000.  Not today.

The Sequent machine looked attractive.  Reasonable performance, a fairly
affordable price, and above all expandability.  As I mentioned in an early
note to this mailing list, we probably can't afford something heavy enough
to keep us happy five years from now.  The ability to plug in more CPUs
was definitely a significant plus for Sequent.  We liked the 32000-series
architecture.  And performance looked higher than some of the other boxes
we'd been considering.  Parallelism per se was not much of an issue; we
are not into parallel-processing research, and to our minds splitting
X MIPS of power into n pieces of X/n MIPS each just means that it's hard
to put all X to work on one problem.  (Not an issue either way in our
environment, which is many-user timesharing, except insofar as X and n*X/n
might have different price tags.)

There were some warts as well, which got bigger and uglier the more we
looked at them.  Having the fast disks on the Multibus wasn't terrific,
although I hear this is now being remedied.  We automatically distrust
designs using custom chips, since that implies being tied to one source
for spares.  (We do not generally do our own CPU maintenance, but with
the supplier being a new and small company, the possibility had to be
faced.  Sequent does look solider now, but our distrust of single-sourced
parts remains.)  We saw parallelism as a potential stumbling block and
source of hard-to-find bugs if we did kernel modifications; a uniprocessor
definitely is simpler to work with.  This was a non-trivial issue since
Sequent uses 4.2, notorious for being buggy and for having made some
really stupid changes that we might want to reverse.  (We realize that
Sequent has certainly done a lot of bug-fixing, but the only way to be
*certain* that bug X gets fixed is to have the capability of doing it
yourself if necessary.  Which leads into the big issue in the next
paragraph...)  And the 32000 series increasingly looks like it has missed
the boat; even the 32332 is too little too late.

Finally, we insist on having sources.  We do not want to be at someone
else's mercy for software maintenance, however nicely those people talk
to us while they're trying to sell us a machine.  (Note a common theme
here, related to our dislike of single-sourced hardware components.)
We've had sources for our system since its very beginning.  We are used
to doing our own software maintenance, we are competent to handle it,
and we like it that way.  Our experiences with binary-only software, and
our friends' experiences with it, have been uniformly bad.  We simply
refuse to spend our time working around bugs that we know how to fix!
We also have local kernel enhancements that we are unwilling to give up.

And for sources, Sequent wants an arm, both legs, and several selected
items of internal anatomy.  The list price for the sources was higher
than the list for the hardware configuration!  (Obviously, we are talking
about *complete* sources; sources for the user-level programs are indeed
cheaper, but much less useful, especially since we can borrow most of
them from our existing source anyway.)  Licensing is not the issue; we
are licensed for practically everything under the sun, including at least
one version of Unix (V8) that Sequent cannot be licensed for.

We are willing to sign, and abide by, any reasonable agreement concerning
non-disclosure and proper protection of sources.  We are not, however, in
a position to pay tens of thousands of dollars for the privilege!  Giving
us sources costs the supplier essentially nothing, assuming that we protect
them properly.  The work is already done, and its fruits are distributed
with every binary-only copy, so recovery of costs is an unconvincing
argument.  There are obviously grounds for concern about whether we will
live up to our promises about non-disclosure, but we have never found the
deepness of someone's wallet to have much correlation with trustworthiness.
We could understand unwillingness to release sources except to carefully
selected customers, even if that didn't include us, but we would expect
the selection to be based on reputation and organizational relationships
with Sequent.  The implication that we cannot be trusted because we're poor
strikes us as peculiar and irrational.  In fact, we find it offensive.

We know that Sequent is willing to negotiate on source prices under some
circumstances, but even a drastic reduction would still represent a serious
increase in the net price of the machine.  If Sequent was our only possible
supplier, we would go with this and probably be happy with it.  But Sequent
is *not* our only possible supplier.

We looked at a number of other possibilities, and considered the Celerity
C1200 in particular to be a serious competitor for Sequent.  We had some
doubts about the Celerity machine's performance on multi-user timesharing,
although it certainly does benchmark impressively on number-crunching.
Celerity was willing to include sources at nominal cost, a major point.

But what really killed our interest in Sequent was when the local Integrated
Solutions distributor presented us with numbers on ISI's 68020 box.  This
machine probably isn't as fast as a full-house Sequent, but it's faster
than any configuration we were likely to have any time soon.  Uniprocessor.
Complete sources at nominal cost.  32-bit bus (VME), with auxiliary bus
for fast (faster than the Sun 3) memory access.  And the killer:  a lower
price for a bigger configuration.  Sic transit gloria Sequenti.

				Henry Spencer @ U of Toronto Zoology
				{allegra,ihnp4,linus,decvax}!utzoo!henry