[37391] in North American Network Operators' Group
Re: 10 gige experience
daemon@ATHENA.MIT.EDU (Greg Maxwell)
Thu May 10 08:37:58 2001
Date: Thu, 10 May 2001 08:31:46 -0400 (EDT)
From: Greg Maxwell <gmaxwell@martin.fl.us>
To: Chris Wedgwood <chris.wedgwood@clear.co.nz>
Cc: Vinay Bannai <bannai@pacbell.net>, Mike Leber <mleber@he.net>,
nanog@merit.edu
In-Reply-To: <20010511002045.A6095@styx.clear.net.nz>
Message-ID: <Pine.GSO.3.96.1010510082209.8845C-100000@da1server>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Errors-To: owner-nanog-outgoing@merit.edu
On Fri, 11 May 2001, Chris Wedgwood wrote:
> Anyone know any good articles on 10GB ethernet? I am specifically
> intrested if they have removed some of the legacy requirements
> that hamper gigabit ethernet (such as a default MTU of 1500,
> half-duplex, etc.)
>
> Not answering your question, but GigE can use larger frames than 1500
> bytes, however, if you wish to send these outside your network,
> you'll likely need to fragment them this somewhat less useful... of
> course, Jumbo Frames are still useful within your network, but then
> again most switches don't support them[1].
Which is why I said 'default' MTU of 1500. I think it should have been a
requirement of all 100TX devices to be able to accept FDDI sized packets,
a requirement of GIGE to switch FDDI sized packets, etc.. Thats not what
happened.
The problem is that there is no graceful way to handle mixed MTU's at that
layer. As a result, the default is to use legacy sized MTUs, which results
in new equpment which can not handle the larger MTUs, resulting in the
further delay of end-to-end jumbogram support.
At some point, they need to draw a line and say that reception and
switching of large >1500 byte packets must be supported and generation of
such packets should be an option. This will finally make is possible for
people to jumbogram enable their networks.
