Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: [10GMMF] TP-3 Stressors



See comment in-line.

John

> -----Original Message-----
> From: owner-stds-802-3-10gmmf@IEEE.ORG [mailto:owner-stds-802-3-
> 10gmmf@IEEE.ORG] On Behalf Of Piers Dawe
> Sent: Tuesday, April 19, 2005 7:19 AM
> To: STDS-802-3-10GMMF@LISTSERV.IEEE.ORG
> Subject: Re: [10GMMF] TP-3 Stressors
> 
> Norm,
> 
> The more you stand back from the detail, the easier it is to understand!
> 
> Item 14 of the PAR says:
> 
> 14. Purpose of Proposed Project:
> This project will define a lower-cost, 10Gb/s serial PHY that supports a
> link distance of at least 220m over installed FDDI-grade multimode fiber.
> The specification should enable migration to smaller form factor pluggable
> modules.
> 
> As I see it, our priorities are:
> 
> Acceptable cost
> Acceptable heat and size possibilities
> Acceptable timescale
> 
> This means we must set the bar to what is practicable and high yielding in
> 2005.  

For me this is exactly where the rub is.  For an extremely fundamental
spec/test ("the bar"), we have a very qualitative and subjective set of
qualifiers to set its limit if one were to follow the logic above.  As has
been discussed at length, "high yielding" is nearly an unknowable as
quantifying the installed base with the level of precision which some are
looking for is challenging at best, if not nearly impossible.  A bit more
troublesome for me personally is "what is practical in 2005" is then AND'ed
with this first item.  As we know, silicon revisions are occurring
constantly, and so my practical today is certainly different than it will be
at the July Plenary, which in turn will be different than when we are
scheduled to complete the standard.  All 3 of these could result in product
which I could deem 'practical' in 2005.  While I certainly understand &
appreciate the time-to-market interdependencies here, I have a difficult
time believing that within 802.3 we would be willing accept anything less
than what it would take to produce a robust specification which in the end
will be judged to meet the needs of the users.  If that means I need to do
some more work on my practical implementation today, so be it.

John

> We don't have to agree whether any limitations are "fundamental".
> We don't (and cannot precisely) tie TP3 testing to coverage in the field.
> We know that for any performance level, somebody somewhere can say "if you
> had made performance a little better my old fiber would have worked with
> LRM".  We should not be diverted by such objective creep because it is a
> journey with no end.
> 
> All, we need to do more to show what is practicable.  In particular, we
> need to prove by experiment that we have a correct and viable combination
> of noise loading and stressors in the TP3 test.  Until we have factored
> noise loading in, we cannot sign off the stressors.
> 
> Piers
> 
> > -----Original Message-----
> > From: owner-stds-802-3-10gmmf@IEEE.ORG
> > [mailto:owner-stds-802-3-10gmmf@IEEE.ORG]On Behalf Of Norman Swenson
> > Sent: 18 April 2005 21:40
> > To: STDS-802-3-10GMMF@LISTSERV.IEEE.ORG
> > Subject: [10GMMF] TP-3 Stressors
> >
> > I would like to better understand the position of those
> > advocating a limit of PIE-D in the range of 3.8-4.0 for the
> > TP-3 stressor selection.  I believe we can make progress on
> > the reflector versus taking up weekly TP-3 call time to
> > debate this issue.
> >
> > So far, as I understand the arguments for a PIE-D of 3.8-4.0,
> > they are:
> >
> > 1)       Real-world fiber will show dynamic variation in IPR
> > that will degrade equalizer performance compared to that
> > measured with a static TP-3 test, which somehow leads to a
> > cliff in real-world performance if we set the PIE-D value too
> > high in the TP-3 stress test
> > 2)       Cost-effective and power-efficient EDC chips can be
> > produced that achieve performance in this range
> >
> > Argument (1) would seem to push for higher performance
> > requirements in the static TP-3 test, since margin must be
> > built into the static test to allow for unmodeled dynamic impairments.
> >
> > Argument (2) has been put forward as though there is some
> > fundamental limitation that prevents cost-effective and
> > power-efficient EDC chips from being produced that can
> > equalize PIE-D values higher than 4.0 dB.  I am very
> > interested in learning more about the theoretical or
> > experimental basis for concluding that cost-effective,
> > power-efficient EDC chips cannot perform beyond the 3.8-4.0
> > dB range that has been proposed as a performance metric.
> >
> > Until recently, the committee has focused on quantitative
> > analysis to determine performance requirements for EDC based
> > on theoretical models and measured data of optical fiber,
> > with an implicit, if not explicit, goal of 99% coverage of
> > 300m OM-1 fibers.  Lately, there has been a shift away from
> > considering the percentage of the installed base that can be
> > covered to setting PIE-D objectives independently of coverage
> > requirements.  I do not understand the significance of PIE-D
> > requirements if they are not put in the context of the
> > percentage of the installed base that the PIE-D corresponds
> > to.  Without such context, the numbers can be arbitrarily
> > chosen without any real-world significance.
> >
> > Thanks to any who can help me better understand the arguments
> > for setting PIE-D limits to such a low value compared to that
> > required to achieve 99% coverage.
> >
> >