From: nmm1@cus.cam.ac.uk (Nick Maclaren)
Newsgroups: comp.parallel
Subject: Re: Myrinet vs Gigabit ethernet
Date: 12 Dec 1998 06:08:52 GMT
Organization: University of Cambridge, England
Approved: bigrigg@cs.cmu.edu
Message-Id: <74t19k$kgi$1@encore.ece.cmu.edu>
Originator: bigrigg@ece.cmu.edu


In article <72oafv$r14$1@encore.ece.cmu.edu>, Lars Rzymianowicz <lr@mufasa.informatik.uni-mannheim.de> writes:
|> Greg Lindahl wrote:
|> > Aakash Kambuj <aakashdk@comp.nus.edu.sg> writes:
|> > > Im currently working on building a distributed computing system for a
|> > > project. I would like to know if I should use Myrinet or gigabit
|> > > ethernet.
|> > Benchmarks will show a huge difference in performance, even though are
|> > supposedly the same speed. The problem is that you can't afford to go
|> > through the kernel at gigabit speeds. Once gigabit ethernet adapters
|> > get a CPU on the card and a user-level interface (such as U-Net or
|> > VIA), they will perform similarly to Myrinet.
|> 
|> A good point. Remember: Gigabit Ethernet is "Ethernet, but only faster".
|> So if you have communication-intensive applications, a standard LAN with
|> its large latencies (100+ us) is not an ideal choice.

And, once we have the Ethernet framing in hardware, all the bugs will
be burnt into silicon and hence won't get fixed - not to say that there
will be a dozen different, undocumented "higher-level" interfaces, one
for each vendor :-(

|> And Gigabit Ethernet is too expensive today. Myrinet is a good choice
|> (especially with optimized software layers on it), which can deliver
|> bandwidths of 70-110 MByte/s at latencies of 10-30 us. (Numbers vary
|> with the protocol layer, the more features you want (security, multi-user,...),
|> the more latency or less BW you get).
|> And costs are in the range of 1200-1500$ per NIC-card, 2000$+ for switches.
|> 
|> But there are some other SANs (System-Area-Networks) around:

Yes.  The current front-runner for portable high-performance computing
is HiPPI and GSN (HiPPI-6400) delivers 800 MB/sec peak (and over 400
measured.)


Regards,
Nick Maclaren,
University of Cambridge Computing Service,
New Museums Site, Pembroke Street, Cambridge CB2 3QG, England.
Email:  nmm1@cam.ac.uk
Tel.:  +44 1223 334761    Fax:  +44 1223 334679

--
Articles to bigrigg+parallel@cs.cmu.edu (Admin: bigrigg@cs.cmu.edu)
Archive: http://www.hensa.ac.uk/parallel/internet/usenet/comp.parallel