Lack of PCI Compliance of PLX's PCI 9054 I/O buffers

["O'Shea, David J" <david.j.oshea@intel.com>]

I think it is soapbox time again.  (Or in this case public
stoning time).

My nasty little tear at PLX was only half deserved
by them.  I am certainly no signal integrety expert,
nor an analog expert.  But I can read the specification.
The people that wrote that part of the specification
were experts, and they put down the requirements there
for good reasons.   All we have is the specification.

Surely, many of the good reasons such requirements
were spec'ed are understood by PLX and were the basis
in some part for their buffer designs, and their compliance
claims.  There are many, many factors that went into
the original requirements.  So just as surely, a device
that does not comply to the strict requirements of the
specification will end up breaking some element of the
good citizenship designed into the specification's requirements.

PLX can run tests from now until forever, and it would
not convince me or most other engineers of the validity
of their compatibility claims.  And short of real strict compliance
with the specification, they should not be making such
claims, no matter how many tests they might have run.
They won't be the first company to claim compliance absent the
truth of it.

Surely this is a difficult topic.  Implementing a universal
device to the actual PCI specification seems nearly impossible.
Each vendor is going off and coming up with their own
possibly good solutions, almost all of which are non-compliant,
and almost all of which have not been tested vigoursly
enough against each other.   Its doubtful that absent the rigourous
theoretic review of specification working groups that any
amount of testint is enough.

The theory behind such ideas has not been
hammered out in any subcommittee of the SIG.  A place
were PLX, Xilinx, LSI, IBM, NEC, Fujitsu, etc. could all get
their guys together and revise the specification.  Revise
the specification in such a way that universal cards that
are designed to it will always work and can always actually
be implemented and still be compliant with the specification.

Instead of going to this considerable effort, the companies
are choosing to spend marketing resouces claiming that
they are compliant, when in truth they clearly are not.
They might work, they might not.  But the doubt lingers so
great that its truly just the biggest form of hubris to claim
compliance.

There are no compliance police in the PCI community.
The design engineers at each and every company are the
compliance police.  The interoperability testing is just a
fair and intelligent double check on the design methodology.
But the specification is the only real source of compliance.

Testing does not replace the methodology, and PLX and
other companies anger people by claiming in the clearest 
and most obvious form of marketing that testing has done just that.   

The sad part of it in PLX's case is that they had a customer
who asked all the right questions.  He just did not get all the
right answers from PLX.   So down the road when this became
clear, he was mad.   Not a major suprise.   Any design engineer
in the same space would be just as furious.  If you don't ask,
you can only blame yourself.  But if you ask and get deceit, anger
then seems like just the correct response.

I am sure that PLX's product reps and field engineers
work tirelessly to help their customers get the job done.  But in
this case, your customer asked a basic, fair, honest question,
and didn't get an honest answer.    PLX clearly needs to work
even harder to make sure that type of thing does not happen.
Nearly all companies need to do the same.  There are few saints
in this industry.  I'm sure my current employer has just a few
skeletons of failure in the closet as well.

But papering over the issue later with claims of compliance
just irritates me (and I am sure lots of other design engineers).
Its the wrong approach to keeping customers.

If you watch closely, you can see me stepping down from the
soapbox now.

-David O'Shea
Corollary Inc.
daveo@corollary.com