Re: Cray to commercialize Red Storm

"Nick Maclaren" <[EMAIL PROTECTED]> wrote in message
news:[EMAIL PROTECTED]
> In article <[EMAIL PROTECTED]>,
> Rick Jones  <[EMAIL PROTECTED]> wrote:
> >Nick Maclaren <[EMAIL PROTECTED]> wrote:
> >> Well, almost.  That was so, then.  But the problems with Fast
> >> Ethernet being CPU bound meant that Gigabit Ethernet cards offloaded
> >> at least framing, and perhaps a bit more.
> >
> >Problems with FE being CPU bound probably depended entirely on the
> >market space.
>
> Not really.  When I first tested it, most systems saturated one CPU
> before saturating the card, and it took 50% of a fastish Alpha just
> to drive it at 80% of capacity.  Of course, CPUs got faster, but
> the problem is at least partly memory latency.

Nick, was that Alpha a pre-byte load/store one running
something like NT 3.51 ? If so I would not be surprised
to find that it struggled to push a fast ethernet card.
I came across such a system and it did have trouble with
saturating fast ethernet. AFAICT it was driving the card
through x86 emulation at some level. Unfortuntely I did
not have enough time or need to investigate it
thoroughly. :(

As others in addition to myself have pointed out we've
seen Fast Ethernet saturated doing real work on things
like Pentium Pros way back when... :)

Cheers,
Rupert