TECH TALK
By Paul Motter (TMF
DotCom)
NEW YORK, NY (May 19, 1997) --
Why Intel Should Continue to Cut Fast Ethernet Prices...
We are witnessing the advent of gigabit ethernet products like top-40 one
hit wonders. Sure it sounds cool, sure it is sexy, but...let me caution investors
not to jump on the inevitable bandwagon of hopefuls expecting the entire
industry to convert to gigabit ethernet products overnight. It ain't gonna
happen.
Gigabit ethernet is fast, too fast for most Intel pentium class servers which
make up the vast majority of servers on the market. Independent testing of
existing ethernet products by several different organizations have come to
this conclusion. Most Pentium class servers can barely keep up with 100 mbps
fast ethernet. Forget about 1000 mbps!
A recent article in Network World (Planning for Gigabit Ethernet By Kevin
Tolly 04/28/97) which conducted an independent test of existing "pre-standard"
gigabit ethernet products, says... "Servers, and even client stations, can
easily generate traffic sufficient to fill a 10M bit/sec Ethernet. But that
isn't the case with 100M bit/sec or faster links. At higher speeds, the network
often is not the bottleneck. In fact, you can expect throughput levels of
well under 100M bit/sec from most Pentium-class servers. Trying to drive
a 155M bit/sec ATM-connected server at or near line rate requires Pentium
Pro hardware."
The author concludes that for the near-term, in Intel-based servers, gigabit
ethernet is over-kill and that the only truly beneficial application will
be in switch to switch connections. But that isn't why I am here.
We have been witnessing the war between INTEL <% if gsSubBrand = "aolsnapshot" then Response.Write("(Nasdaq: INTC)") else Response.Write("(Nasdaq: INTC)") end if %> and
3COM <% if gsSubBrand = "aolsnapshot" then Response.Write("(Nasdaq: COMS)") else Response.Write("(Nasdaq: COMS)") end if %> the last few months. 3Com is in the business of
selling LAN switches and hubs, some WAN hardware, and the management software
to go with it. The company makes great products, and has even emerged lately
as a leader in ATM technology for the LAN. But NICs (network interface cards)
make up a substantial part of their business they have relied upon in the
past for steady income.
But Intel is also a maker of LAN hardware, particularly 10/100 ethernet NICs,
and they have been steadily cutting prices. Doesn't Intel care about margins?
Don't they need to make money on the NICs they make, too? The answer is...
NO.
Think about the findings of the gigabit ethernet tests above. The typical
Pentium server cannot fully utilize the bandwidth offered by fast ethernet,
it requires a Pentium Pro. Intel makes its money from selling processors...
millions and millions of them. What they need is incentive for people to
buy new processors, to upgrade what they already have. These days new models
of Pentium come along faster than customers can keep up with, and a new model
processor is simply not the incentive to upgrade it used to be. What does
Intel need? Exactly... incentive. Fast ethernet is one of the best reasons
any customer has to upgrade to a Pentium Pro right now, and fast ethernet
is catching on. Intel has finally decided that if it is going to sell new
processors it has to give customers a reason to want them. That is why they
are exploring video-conferencing, multi-media, gigabit ethernet, and other
processor intensive applications. By cutting prices on NICs they not only
sell more processors but they also do something else they are famous for...
They trounce the competition. |