RE: [10GBASE-T] PAR and 5 critters
In deference to some of Brad & Bob Grow's earlier admonition, technical
feasibility is a matter of increasing confidence as time moves on. The
tone of this discussion appears to have moved from the "can't be done at
all" to "how much & what kind of silicon will it require". I will assume
that we have entered that stage.
We have presented our estimates of feasibility at about 6X 1000 BASE-T
and implementable in today's CMOS at the tutorial in November.
Regardless, there is no doubt that 90nm will be a commercial processes
well before 10GBASE-T is through the standards process (at the earliest
2nd half of 2005), and 65nm will be commercial as 10GBASE-T begins to
ramp in the subsequent years.
In direct response to Dan's concern, there are a variety of algorithms
that do not require closing the loop at the baud rate, (the simplest of
which are the look-ahead algorithms which have obvious complexity
drawbacks), various reduced-state and lower-complexity forms are well
studied in the literature, and have been implemented in commercial
products. (Dan - you will also see EMI measurements from November)
In deference to earlier comments by Vivek & others, yes, if I just take
the simplest form of design (direct-form FIR) and multiply up by the
baud rate & # of taps I get a huge complexity multipler (something like
the 45X 1000 BASE-T), but just because the simplest-extension yields a
huge complexity doesn't mean that it is non-feasible. Current art in
efficient and multi-rate filtering algorithms don't scale linearly as
the number of taps or processing speeds go up, and provide significant
benefits (see November tutorial for an example, but these are also well
studied in the literature).
George Zimmerman
gzimmerman@solarflare.com
tel: (949) 581-6830 ext. 2500
cell: (310) 920-3860
> -----Original Message-----
> From: DOVE,DANIEL J (HP-Roseville,ex1) [mailto:dan.dove@hp.com]
> Sent: Wednesday, July 30, 2003 8:38 AM
> To: 'sreen@vativ.com'; 'Alan Flatman'; 'Kardontchik, Jaime'
> Cc: '[unknown]'; 'Sterling Vaden'
> Subject: RE: [10GBASE-T] PAR and 5 critters
>
>
> > 2. Performing large number (x8 relative to 1000BaseT) of DSP
> > calculations at
> > 833MHz,
>
> Precisely my concern. I think it would be a useful exercise to
> calculate the loop timing necessary to make such a thing work, and
> then extrapolate to the process geometry that would enable it.
>
> I observed that 1000BASE-T did not really become solid in practice
> until .18u became available. There were some decent .25 designs, but
> I suspect that corners were being "trimmed" to make timing close.
>
> Dan
>
> > -----Original Message-----
> > From: Sreen Raghavan [mailto:sreen-raghavan@vativ.com]
> > Sent: Tuesday, July 29, 2003 1:58 PM
> > To: 'DOVE,DANIEL J (HP-Roseville,ex1)'; sreen@vativ.com;
> > 'Alan Flatman';
> > 'Kardontchik, Jaime'
> > Cc: '[unknown]'; 'Sterling Vaden'
> > Subject: RE: [10GBASE-T] PAR and 5 critters
> >
> >
> > Dan:
> >
> > We are really referring to the theory (Shannon Capacity) when
> > we say 10Gbps
> > cannot be achieved over CAT-5e or CAT-6 cabling. Theory shows
> > that 10Gbps
> > can be achieved over CAT-7 cabling. Practical issues to
> > accomplish 10Gbps
> > over CAT-7 cabling include (assuming PAM-10 modulation):
> >
> > 1. Building an 11-bit effective ADC at 833 MBaud,
> > 2. Performing large number (x8 relative to 1000BaseT) of DSP
> > calculations at
> > 833MHz,
> > 3. DDFSE critical path to be implemented in 1.2 ns
> > 4. Building a linear transmit driver with an 833MGz bandwidth
> > & 40 dB SNR
> >
> > The above list by no means is exhaustive, but shows the
implementation
> > issues that need to be considered.
> >
> > Sreen
> >
> > -----Original Message-----
> > From: DOVE,DANIEL J (HP-Roseville,ex1) [mailto:dan.dove@hp.com]
> > Sent: Tuesday, July 29, 2003 1:09 PM
> > To: 'sreen@vativ.com'; 'Alan Flatman'; 'Kardontchik, Jaime'
> > Cc: '[unknown]'; 'Sterling Vaden'
> > Subject: RE: [10GBASE-T] PAR and 5 critters
> >
> > Hi Sreen,
> >
> > One thing that occurs to me on this point is the difference between
> > theory and application. Specifically, how many process actions have
to
> > take place within a baud time to close the loops on the DSP and what
> > process geometry would be required to make that timing closure?
> >
> > I know that with 1000BASE-T, the theory was rock solid long before
the
> > processes to implement it were reliable.
> >
> > Dan
> > HP ProCurve
> >
> > > -----Original Message-----
> > > From: Sreen Raghavan [mailto:sreen-raghavan@vativ.com]
> > > Sent: Tuesday, July 29, 2003 11:52 AM
> > > To: 'Alan Flatman'; 'Kardontchik, Jaime'
> > > Cc: '[unknown]'; 'Sterling Vaden'
> > > Subject: RE: [10GBASE-T] PAR and 5 critters
> > >
> > >
> > >
> > > Just to clarify, Vativ, Broadcom & Marvell presented capacity
> > > calculations
> > > at the Portsmouth meeting and showed that worst-case CAT-7
> > > (Class F) cabling
> > > had sufficient channel capacity to achieve 10Gbps throughput
> > > at 100 meter
> > > distance. The reason for "may be possible" statement in the
> > > conclusions was
> > > that the 3 PHY vendors felt that more work needed to be done
> > > on practical
> > > implementation issues before the conclusion could be
> > altered to a more
> > > definitive statement.
> > >
> > > In addition, we proved conclusively that there was NOT
> > > sufficient channel
> > > capacity on existing CAT-5e (Class D), or CAT-6 (Class E)
> > > cables to achieve
> > > 10 Gbps throughput.
> > >
> > > Sreen Raghavan
> > > Vativ Technologies
> > >
> > > -----Original Message-----
> > > From: owner-stds-802-3-10gbt@majordomo.ieee.org
> > > [mailto:owner-stds-802-3-10gbt@majordomo.ieee.org] On Behalf
> > > Of Alan Flatman
> > > Sent: Tuesday, July 29, 2003 9:51 AM
> > > To: Kardontchik, Jaime
> > > Cc: [unknown]; Sterling Vaden
> > > Subject: RE: [10GBASE-T] PAR and 5 critters
> > >
> > >
> > > Message text written by "Kardontchik, Jaime"
> > > >Was any reason given why it would not run on Class F ? Was it for
> > > technical reasons or for marketing reasons ?<
> > >
> > > The 3-PHY vendor presentation made in Portsmouth (sallaway_1_0503)
> > > calculated 49.36 Gbit/s capacity using unscaled Cat 7/Class F
> > > cabling. This
> > > figure was reduced to 37.71 Gbit/s with worst case limits.
> > Overall, I
> > > thought that this was a refreshingly realistic presentation and I
> > > interpreted the summary statement "Capacity calculations with
> > > measured data
> > > indicate 10 Gigabit data transmission over 100m Cat 7 may
> > be possible"
> > > (slide 16, bullet 3) as overly cautious engineering judgement.
> > >
> > > So, what has changed since the May interim? Not the laws of
physics!
> > >
> > > Best regards,
> > >
> > > Alan Flatman
> > > Principal Consultant
> > > LAN Technologies
> > >
> > >
> > >
> > >
> > >
> >
> >