FNN vs GigabitEther & Myrinet
bogdan.costescu at iwr.uni-heidelberg.de
Thu Oct 25 15:17:09 PDT 2001
On Thu, 25 Oct 2001, Patrick Geoffray wrote:
> Donald Becker wrote:
> > GAMMA to be widely adopted -- we could put the proper hooks into the
> > drivers. But that's a large effort that would be a resource and
> > financial drain, without any associated revenue.
> Well, if your product can provide, for a medium/large panel of
> Ethernet and GigE NICs, low latency and OS-Bypass protocols (socket+MPI),
> the market for Ethernet based clusters is definitively yours.
Well, I was only thinking about MPI, no sockets - this might be more
difficult, as you still have to follow the TCP/IP specifications. OTOH,
MPI is a high enough level protocol to be able to run on top of something
which doesn't have to follow any such rules.
My initial line of thoughts was not to have GAMMA (or something similar)
able to hook into the Scyld's drivers, but to have it _use_ the Scyld
drivers. This way both directions work: the low-latency package can
have the driver coverage provided by Scyld's knowledge and Scyld can have
a low-latency package which is guaranteed to work with their drivers.
Of course, this implies some kind of involvment from Scyld in the
low-latency package which might not be desired by the low-latency package
initial authors (selfishness, fear to loose the development direction,
etc.) and/or by Scyld (not enough resources, etc.).
So, my second line of thoughts was that Scyld might develop something on
their own - either open-source or partially open-source (the kernel part
probably has to be GPL, anyway). As I said - they have the knowledge, and
the fact that it didn't happen until now suggested that either they don't
have the resources or that, for some other reason, they don't want to do
it. Don's answer cleared my suspicions...
> Now, can it be a source of substancial revenue ? Not simple
> if the soft is completely open-source.
I don't think that this is different from the current situation. All the
software that Scyld provides is open-source, still they do get revenues.
Another possibility would be to have part of the software being sold: for
example the MPI library on top of the low-latency communication protocol;
there could be an open-source MPI library and a proprietary one
(supposedly better optimized, etc.) - like MPICH/LAM-MPI vs. MPI-Pro.
Any takers ?
The above thoughts might be regarded as gratuitous publicity. So, I have
to say that:
I don't work for Scyld, I'm not paid by Scyld and I have no other
connection with them than good colaboration with Don w.r.t. driver
development (3c59x for those who don't know) and associated utilities.
I would be equally happy if some other company would fill this gap.
IWR - Interdisziplinaeres Zentrum fuer Wissenschaftliches Rechnen
Universitaet Heidelberg, INF 368, D-69120 Heidelberg, GERMANY
Telephone: +49 6221 54 8869, Telefax: +49 6221 54 8868
E-mail: Bogdan.Costescu at IWR.Uni-Heidelberg.De
More information about the Beowulf