RE: Some Gigabit Ethernet history -- why are the links so overdesigned?
- To: "Joe Gwinn" <gwinn@xxxxxxxxxx>, <stds-802-3-hssg@xxxxxxxxxxxxxxxxxx>
- Subject: RE: Some Gigabit Ethernet history -- why are the links so overdesigned?
- From: "Michael M. Salzman" <msalzman@xxxxxxx>
- Date: Thu, 13 May 1999 21:25:13 -0700
- Importance: Normal
- In-Reply-To: <firstname.lastname@example.org>
- Reply-To: <msalzman@xxxxxxxxxx>
- Sender: owner-stds-802-3-hssg@xxxxxxxxxxxxxxxxxx
Joe, your recollections are partially correct. It was not just variability
but certain impairments that were discovered upon closer examination. In
any case, the installed base is just that.
It appears that the CWDM approach or the MAS approach may support 10Gig over
the installed base - why abandon it? Is this a case of "real engineers
don't do the installed base?" Why not simply specify a workable solution
to the problem as the 10G-IBSX, ie. grungy Installed Base X Phy.
We can then move on and design a really cool, serial ethernet, block encoded
phy, for real engineers.
Just kidding folks. But seriously what is wrong with a multi phy/pmd
approach. I take it as a foregone conclusion.
Perhaps the issue you are raising is the parable of 1000BaseT, which first
agonized about taking the 10B output, and decoding it first, in order to
derive the original 8B, in order to then encode it properly for the twisted
pair phy. We may have to part company with each other at the 10GMII level,
rather than at the PMA level, which would complicate life for some of us
perhaps. Life is hard, but I dont' think we can make it easier by
abandoning the installed base.
The market growth we seek will come with casual use, and a PHY that requires
only "really good fiber" cannot be a casual use solution.
> -----Original Message-----
> From: owner-stds-802-3-hssg@xxxxxxxxxxxxxxxxxx
> [mailto:owner-stds-802-3-hssg@xxxxxxxxxxxxxxxxxx]On Behalf Of Joe Gwinn
> Sent: Thursday, May 13, 1999 08:02
> To: Paul Bottorff; bill.st.arnaud; Ed Grivna
> Cc: stds-802-3-hssg
> Subject: Some Gigabit Ethernet history -- why are the links so
> At 12:00 PM 99/5/11, Paul Bottorff wrote:
> >In the LAN code efficiency is less important since cable is free and
> >amplifiers are not required. Never the less 1 Gigabit had great trouble
> >making the 500 m spec for multimode fiber. Reductions in reach
> outside the
> >standards for building wiring are expensive for LAN applications.
> There is a little bit of Gigabit Ethernet history that may be useful here:
> If you recall from the GbE Modal Bandwidth Investigation (MBI) effort, the
> key issue in setting the guaranteed link lengths was the wide variability
> in fiber characteristics (with uncontrolled laser launches),
> especially for
> the dusty legacy fibers of the installed base. Because the then objective
> was to ensure that at least 99% of conforming links would work with
> randomly chosen transceivers, the 1% tail of the fiber bandwidth
> distribution wagged the dog, resulting in grossly overdesigned
> (~4:1) links
> in the typical case. So, if we can find a way to reduce the variability,
> even if the average bandwidth is unchanged, we may be able to
> increase the guaranteed lengths.
> I would suggest that before there is much discussion of possible coding
> schemes, the committee should first debate and decide if now is
> the time to
> cut the legacy cord, or not.
> If the cord is to be cut, it's best done cleanly.
> Joe Gwinn