fast blackplanes (was Re: Experience with SMC 8624T L2 switch es and nics)

Marc Cozzi cozzi at nd.edu
Wed Feb 26 09:13:43 PST 2003


Dale,

For me that is the point of concern. A "fast backplane" switch such as
FastIron/BigIron or even the Cisco 6509 series would be nice but the price
for a 100 node cluster could be between 70 and 100K US dollars.

If you are only using 24 or less ports, finding a non-blocking switch that
can handle 48Gb/s is not really an expensive problem.

Trunking capabilities in the SMC switch could allow one to
in-expensively construct a cluster with many more nodes with overall
performance better than simple switches fitted with GBICs.

For a cluster much larger than say 48-64 nodes I don't think you could out
perform the "fast backplane" type switches, however, the SMC tiger seems
to fill a nitch somewhere in between the two.


  marc


-----Original Message-----
From: Dale Harris [mailto:rodmur at maybe.org]
Sent: February 25, 2003 5:14 PM
To: 'beowulf at beowulf.org'
Subject: fast blackplanes (was Re: Experience with SMC 8624T L2 switches
and nics)


On Tue, Feb 25, 2003 at 08:42:47AM -0500, Mark Hahn elucidated:
> 
> it is a good deal.  I know of a large cluster recently installed
> in Toronto that is very happily using these switches in an agressive
> way (somewhat exotic wiring plan, including trunking.)  I believe
> they are quite happy with performance.  we have one which I'm 
> only starting to evaluate the performance of, but so far, so good.
> 


I guess I'm missing something here.  How would this compare to going
with a switch that has a large number of ports with a really fast
backplane?  Like say a Foundry BigIron 4000, or some such.  Is this a
just a cost vs. performance issue?  Can these switches be wired in such
a way to out perform a fast backplane?


-- 
Dale Harris   
rodmur at maybe.org
/.-)
_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list