MPI over ATM help or how-to?

Gerry Creager N5JXS gerry at
Sat Mar 30 18:11:34 EST 2002

Marketting types, setting cost-per-port for ATM at 4x GBE and 20x
100BaseT effectively killed ATM for LAN and SAN.  It's still viable for
operations that benefit from connection-oriented operations, in the case
of this forum, that being the Grid initiatives.

I don't see ATM returning to the LAN forefront.  I see GBE to the
desktop and MPLS-based protocols surging ahead of it.  That said, I
suspect that after a foray into packet-based protocols without ensured
sessions connectivity the carriers will return to ATM or MPLS with
DiffServ-based QoS.

But then again, I'm really an ATM bigot...

joachim at wrote:
> > On Fri, Mar 29, 2002 at 09:13:19AM -0500, Lechner, David wrote:
> >
> > > I am trying to do a comparison for customers that use ATM now - our goal is
> > > to demonstrate the performance improvement in running native and optimized
> > > FE communications vs. ATM (the cost savings is obvious) -
> >
> > I looked on the web and the only "native" ATM MPI implementation I see
> > is MetaMPICH, which seems to have gone defunct before delivering the
> > ATM part. So you are stuck with IP emulation over ATM, which is
> > probably never faster than fast ethernet.
> MetaMPICH (which I was involved with) was designed to couple
> distributed MPI machines (T3E and SP2, in this case) via WAN
> (4 x 622Mbit ATM, in this case). Therefore, we needed to use
> multiple message routers on each machine. This worked quite well
> for certain classes of applications (coupled simulations).
> It is no "native"
> ATM usage in the sense that two processes directly exchange
> data via ATM. ATM for LAN or even SAN usage is quite dead, as
> I see it.
>   Joachim
> _______________________________________________
> Beowulf mailing list, Beowulf at
> To change your subscription (digest mode or unsubscribe) visit

Gerry Creager -- gerry at
Network Engineering
Academy for Advanced Telecommunications and Learning Technologies		
Texas A&M University	979.458.4020  (Phone) -- 979.847.8578  (Fax)
Beowulf mailing list, Beowulf at
To change your subscription (digest mode or unsubscribe) visit

More information about the Beowulf mailing list