[Beowulf] The move to gigabit - technical questions

[Jim Lux]

----- Original Message -----
From: "Vincent Diepeveen" <diep at xs4all.nl>
To: "Glen Gardner" <Glen.Gardner at verizon.net>
Cc: <beowulf at beowulf.org>; <rgb at phy.duke.edu>
Sent: Tuesday, March 15, 2005 12:11 PM
Subject: Re: [Beowulf] The move to gigabit - technical questions


> At 05:41 PM 3/14/2005 -0500, Glen Gardner wrote:
> >Gigabit will be a little faster than 100Mbit on a small cluster, but not
> >a lot.
>
> What is 'not a lot'.
>
> I would guess it's factor 10 faster in bandwidth?

I would guess it's not 10 times faster (leaving aside latency and bus
bandwidth to the interface issues). I don't know about the details, but it's
real common to have some sort of synchronization/equalization sequence at
the front of the packet that runs at a lower bit rate than the payload rate.
Wired "thicknet" ethernet, as I recall, had a 64 bit alternating 1/0
preamble before the actual packet contents w/header.  It could be adequately
modeled, though, as some sort of fixed per packet overhead time.

This is especially true for wireless LANs (I know that not many clusters use
these, but as they get faster, and there's more channels available, it gets
attractive.. no cables!).  802.11a/b/g always starts at 2 Mb/sec for the
preamble, and then shifts to a faster modulation (depending on the
propagation).

Jim Lux