Archives


- Beowulf
- Beowulf Announce
- Scyld-users
- Beowulf on Debian

[Beowulf] interconnect and compiler ?

Many of your questions may have already been answered in earlier discussions or in the FAQ. The search results page will indicate current discussions as well as past list serves, articles, and papers.

Search

Mark Hahn hahn at mcmaster.ca
Thu Jan 29 07:28:39 PST 2009


> Can anyone tell me if I can find a complete list of all high BW, low latency 
> networks and vendors somewhere. I know myricom, quadrics and the various 
> infiniband vendors like melanox, verari, voltaire and Qlogic. Are these the 
> main players?

well, do you count 10GE in general?  I would think you should, since 
although 10GE+OpenMP isn't _that_ competitive with IB/quadrics or 
even myriMX, it's still very fast.  the 10GE market seems to be 
developing slower than I expected, though, with switches remaining 
pretty expensive, and there doesn't seem to be a lot of "chatter"
of street experience on this list, for instance.  my organization has had 
a 10GE WAN backbone for a while, and has had moderate success with 
getting 10GE data onto it (rather than just aggregating lots of 1GE).
SFP+ seems like a useful advance for 10GE, assuming 10GbaseT is either
not going to happen or will take a long time.  the switch side of things
seems to be slo-mo as well.

IB is, of course, mainly just Mellanox, though Qlogic still has a finger
in the pie.  (just NICs, though, right?  AFAIKT Qlogic's switches are built
from Mellanox chips like all the rest...)

Quadrics is a bit of an enigma to me - we're still going strong with our 
Elan4 systems, but I don't really understand Quadrics future plans.
they seem to be morphing into a 10GE vendor, but do they have any unique
value-add?

> BTW, as for the adaptors, does it make a big difference if they are on a 
> hypertransport or PCI express (because I might be constrained to Xeon's)?

iirc, there was only a small latency difference between HTX and PCI-E
versions of Infinipath.

> Next, could anyone point me to recent benchmarks of the pathscale and icc 
> compiler vs. gcc ?

I'd like to see that too!  might as well throw in PGI.  in fact, we have 
all of those compilers available - does anyone have suggestions of good 
test codes?

regards, mark hahn.



More information about the Beowulf mailing list