Cheap, good tulips...

[Robert G. Brown]

On Thu, 7 Jan 1999, alex wrote:

> I also wrote to Wisecom, and I got a response.  Unfortunately, I think
> they answered me because they thought I was building a cluster.  
> 
> I tried to explain the linux NIC driver situation, about how the good
> drivers were coming out of NASA, which was developing the drivers for
> their cluster.  I said that if they could make clean, genuine DEC
> tulips available on the web at a low price, they'd arguably be
> offering the best NIC for the buck for linux users, and they'd
> probably sell a lot of cards.
> 
> The response I got asked me about the cluster I was building, so I
> didn't follow up on it.

I'd be patient with them.  One thing a lot of folks on this list don't
seem to have any appreciation of is that these smaller hardware
manufacturers have low capitalization, intense competition, and low
margins.  They cannot afford to properly advertise so their products
need to sell primarily on price point and availability.  They also don't
have a "real" sales department with lots of smart people -- they have a
few channel markets and primarily deal with resellers looking for a
bargain.

It is a very reasonable bet that the higher management of this company
has never heard of linux (or rather, only recently heard of linux when
the Intel/Oracle/Red Hat deal brought linux to the national business
consciousness).  Their business strategy, not unreasonably, has
undoubtedly been to build cheap cards that work with canned software
drivers.  I doubt that they have much of a software shop at all -- the
windows drivers that came with the card were the SAME drivers on the
SAME generic driver floppy that came with several other generic tulip
cards I've gotten in the past.  I would guess that Digital (now Intel)
sells the chip itself and the driver as a package, or perhaps there is a
company that sells the generic driver to OEM's.

Since the card works with the standard tulip driver that comes with
linux (pretty much any version) it isn't overwhelmingly important to me
that wisecom put a linux driver on its floppy, and if wisecom indeed is
buying these floppies mass produced by somebody else for generic tulip
cards, it is basically infinitely unlikely that they will increase their
marginal cost by pushing for the issue.  It would be nice if their
website ACCURATELY indicated card resellers (or had a retail channel for
the cards, although the resellers might not like it if they had such a
thing).  It also might be nice to convince them to put "linux certified"
or some such on the box, if there was any way to "certify" their card --
this would be a nice advertisement both ways.  I don't feel like
expending the necessary energy to bug them about this, but anybody who
ends up dealing with them (when trying to arrange a purchase of a block
of 50 cards or so for a beowulf, for example) might gently make the
suggestions.  Just remember that they want to make money, but they
probably have very little budget for customization or change -- try to
think of ways for them to emphasize linux-friendliness that won't cost
them anything.  A mention of linux compatibility on their website, for
example, is zero-marginal-cost...

> Unfortunately, I think that Intel NICs are probably the way to go.  I
> have some linksys NICs that are working perfectly with the tulip
> driver, but I'm wary of buying more, because you never know what chips
> you'll get.  On top of that, the time we're all putting into this is
> probably worth a lot more than the marginal cost of the Intel cards.

I disagree about the marginal cost -- rather, say that it depends on
your situation.  Intel eepro100 cards cost (correct me if I'm wrong)
$50-70 each, depending on the marketing channel (over the counter costs
are in the $70 category, mail order is a bit cheaper).  Now I agree that
IF one has to spend days finding a usable tulip and are only buying a
few cards, eepros, especially by mail, are a reasonable deal (and work
very well).  I've bought quite a few myself under those circumstances.

On the other hand, if one is fortunate enough to have a local dealer (or
find a reliable mail order outlet) for a tulip card that actually works
perfectly (like the wisecom card or one of the other three or four cards
that folks have just written in about) and they cost $25-30 (the typical
price point retail) there is a clear and overwhelming advantage in
getting them instead.  Face it, this is half the cost per card -- even
if one is buying only one or two cards, the marginal savings is worth an
hour or two of most folks time.  If one is buying for an "operation" (a
network or beowulf) and need 10 or more cards, it is worth spending a
day or two running down the tulips rather than buying eepro's however
convenient the latter.  (Unless, of course, you make several hundred
thousand a year so the $250 saved is smaller than the cost of a day or
two of your time...;-)

It is here that having a list of reliable tulip cards becomes
extremely valuable and important -- it can help SAVE a lot of time and
frustration as one surfs the web retailers in search of a reseller.

> It's too bad there aren't better industry guidelines for labeling
> cards.  I wouldn't mind companies using different chipsets if it was
> possible to cope with it.  Right now, you pretty much have to order
> the card, open up the box, and look at the chips to see what's going
> on.

Totally agreed.  Even the wisecom cards, I always open the box and look,
although if one learns the part number from the website that ought to
suffice as well.  This is also a reason to work, however gently or
lazily, on making the companies that DO make linux-compatible tulips
aware of the fact that there is a market that cares to discourage them
from casually switching chipsets.  Since they provide the Windows
drivers either way, they are inclined to say what does it matter, the
consumer will never know the difference or care.  We need to educate
them on this -- there is a class of consumer that knows and cares very
much, and even though we may be only 5-10% of their market, they are low
margin and that 5-10% translates into perhaps 10-20% of their final
profit.

    rgb

Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb@phy.duke.edu