[Beowulf] Improving access to a Linux beowulf cluster for Windows users

[Joe Landman]

Kilian CAVALOTTI wrote:

> The same benchmark on lower-end hardware (E5345) running Linux (same 4X 
> DDR IB though), gives roughly 30% better results:

While these ping results are interesting, a real app they are not.

I prefer fluent data, sitting on fluent's web site.  Turns out they do 
have identical hardware running Linux and CCS.  I leave it as an 
exercise to the reader to do the analysis.

The startup time claims were interesting, in light of the recent mvapich 
announcement of thousands of nodes booting using their new technology in 
  times small compared to what we saw Microsoft announce.

That said, I would expect that OpenMPI and mvapich on Windows should 
have comparible startup times to Linux.

There are other issues for Windows codes (the 32 -> 64 bit transition is 
one ... been through that at SGI, very hard to convince ISVs to do the 
ports).  Some will reduce performance.  Its a CBA as Jim Lux and RGB 
will tell us ... if the performance difference isn't expensive, it may 
be fine for the user.

I am trying to get started playing with SUA at some point.  SFU wasn't 
all that good to run code in.  I want to see if SUA is any better.  Of 
course I need a machine and install to try this on.  Sadly the way my 
time is being consumed these days, it may be someone else that gets to 
play with these things.

-- 
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics LLC,
email: landman at scalableinformatics.com
web  : http://www.scalableinformatics.com
        http://jackrabbit.scalableinformatics.com
phone: +1 734 786 8423
fax  : +1 866 888 3112
cell : +1 734 612 4615