Re: Jumbo Frames in 10GbE? (Jain's FDDI estimate of CRC max length)

[Ariel Hendel <Ariel.Hendel@Eng.sun.com>]

> X-Sender: gwinn@tif531.res.ray.com
> Mime-Version: 1.0
> Date: Thu, 24 Jun 1999 20:55:34 -0500
> To: ted@alteon.com (Ted Schroeder), bbooth@level1.com, l_d_miller@email.msn.com
> From: gwinn@ed.ray.com (Joe Gwinn)
> Subject: Re: Jumbo Frames in 10GbE?  (Jain's FDDI estimate of CRC max length)
> Cc: stds-802-3-hssg@ieee.org
> X-Resent-To: Multiple Recipients <stds-802-3-hssg@majordomo.ieee.org>
> X-Listname: stds-802-3-hssg
> X-Info: [Un]Subscribe requests to  majordomo@majordomo.ieee.org
> X-Moderator-Address: stds-802-3-hssg-approval@majordomo.ieee.org
> 
> 
> At 1:41 PM 99/6/21, Ted Schroeder wrote:
> >Please check the document Error Characteristics of FDDI by Raj Jain.
> >
> >http://www.netlab.ohio-state.edu/~jain/papers/xie1.htm
> >
> >The CRC-32 algorithm breaks down at 11454 octets given the error
> >characteristics he analyzes in his paper.
> 
> I just read Jain's paper.  His rationale is very FDDI specific, depending
> on the interaction of NRZI with the 4B/5B code used by FDDI, so one would
> expect to get a much different answer for 8B/10B.  My instinct is that the
> Fibre-Channel/GbE answer will be that breakdown occurs only for much larger
> packets than 11,454 bytes, because there is so much more codespace in the
> 8B/10B code than the 4B/5B code.  The breakdown will not occur for shorter
> packets.
> 
> However, as I haven't redone Jain's analysis for 8B/10B, I don't know the
> new breakdown size.  It would be interesting; maybe someday.
> 
> Joe Gwinn
> 
> 

The weakening of the CRC check has to do with the Hamming Distance vs.
Frame Size of the CRC polynomial (table 7 in Jain's paper). Different
coding might make this weakening irrelevant in the big scheme of
things, but it will not change the strength of the polynomial vs.
packet size.

Furthermore, coding arguments cannot help us for packet length because
the longer packets transcend a single link. Whenever a Jumbo frame,
sourced on an 8B/10B link, makes it to a 100BASE-T link it will coded in
4B/5B pretty much like Jain's case.

Ariel Hendel
Sun Microsystems