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

Re: Does only decimal system work for data rate specification?




Bill,

I have been missing you at the 802.3 and HSSG meetings since June, when we
started the WAN PHY debate.  Perhaps you can join us at the Dallas Interim
meeting in January.  I am looking forward to talking to you person to person. I
have been told by several people that you have some very interesting experience
with building Native IP optical networks with CWDM.

Thank you,
Roy Bynum

"Bill St. Arnaud" wrote:

> Your comments on fixed decimal rate is bang on.
>
> I think the question is largely irrelevant of whether long haul 10GbE should
> be clocked at SONET speeds 9.98.. or exactly 10Gbps.
>
> We believe that data variable 10GbE is critical, particularly for CWDM
> systems.  In the CWDM system we are deploying in Alberta for example the
> critical limiting factor is the distance between existing repeater huts.
> The last thing we want to do is build new repeater huts.  A repeater hut is
> a hell lot more expensive than any optical technology.  The problem with
> CWDM is that you have different attenuation and dispersion rates at
> different frequencies and so with fixed clock rates the link loss budget at
> the frequency with the worst attenuation and dispersion (usually around
> 1310) drives your network design. But with variable data rate 10Gbe we could
> lower the clock rate at those frequencies to achieve the same BER.  More
> importantly at frequencies that exceed the link loss budget the clock rate
> could possibly be increased to 14, 16, 18 , 20 Gbps.  This is a big
> advantage of CWDM over DWDM as you don't have to worry about you modulation
> skirt interfering with other wavelengths or being cut off by the DWDM
> filters.
>
> I understand the problem with frequency agile PLLs, but what I had in mind
> is perhaps a GBIC like PHY which we could plug into the 10Gbe card.  I
> believe the proposed 10Gbe standard will have a simple flow control
> mechanism to allow for variable data rates.
>
> I keep reminding vendors that we don't feel a standard for long haul 10GbE
> is that critical to us.  Long haul 10GbE will have the same problems as
> SONET transport and will require both ends to be carefully matched to
> optimize performance and therefore most likely will require both ends to
> come from the same manufacturer.  However the interconnection to the  long
> haul segment should be along the lines recently proposed by the OIF for a 12
> channel  inter-equipment 10GbE.  That is very important.
>
> For LAN based 10Gbe the problem is more complex, because you definitely want
> interoperability between equipment vendors, but I think a variable rate
> 10Gbe would still be useful.
>
> Bill
>
> Bill St. Arnaud
> Senior Director Network Projects
> CANARIE
> bill.st.arnaud@xxxxxxxxxx
> +1 613 785-0426
>
> > -----Original Message-----
> > From: owner-stds-802-3-hssg@xxxxxxxx
> > [mailto:owner-stds-802-3-hssg@xxxxxxxx]On Behalf Of Fenghao Mu
> > Sent: December 9, 1999 3:56 AM
> > To: stds-802-3-hssg@xxxxxxxx
> > Cc: Fenghao Mu
> > Subject: Does only decimal system work for data rate specification?
> >
> >
> >
> > Hello,
> >
> > I have seen some of the emails, and also missed some.
> > I am a designer for a single-chip solution of gigabit Ethernet switcher.
> > I was a little bit confused by this discussion group. I have the
> > following
> > questions:
> >
> > 1) Why we prefer to only specify 10GE rather than suggest other
> > reasonable data rates related to gigabit Ethernet, say  2.5/5 gigabit
> > Ethernet?
> > 2) Why we need Hari coding to tie four individual links together instead
> > of sending data separately?
> > 3) Does our data rate specification fit with market requirement and
> > technology progress well?
> > I think the last question is very critical. If a specification does not
> > match the market and technology progress, it will die!
> >
> > I also have the following points:
> > 1. From a customer point of view, only 1GE and 10GE give less choices
> > for one to build a good switching structure. He has to chose between two
> > possible data rates, either 1GE or 10GE. It might be very
> > difficult for him
> > to upgrade his data links, in other words, 10GE is too luxurious and 1GE
> > is insufficient for his requirement. What is his best solution? Perhaps
> > he has to wait until the price of 10GE comes down to an acceptable value.
> > This acceptable value may never come if the standard does not fit to the
> > market requirement. However, if there is a 2.5/5 gigabit Ethernet
> > data rate,
> >
> > he could flexibly upgrade his network according to his real requirement.
> >
> > 2. Data rate evenly distributed networks are seldom met in
> > reality, and it
> > exists only in people's imagination. Designing a switcher fabric based on
> > this imagination will cause a big waste of resource. For data
> > rate unevenly
> > distributed networks, multiple data rates can give one more solutions,
> > therefore a relative optimum is easy to achieve. The future 10GE switcher
> > will face a serious challenge of limited I/O capacity. How to efficiently
> > use
> > of the I/O pads is a very important issue. If we could use
> > 2.5/5/10 GE data
> > rates, we can individually assign a data rate for a path in a network
> > according different requirements. Multiple data rates can be very helpful
> > for
> > optimal I/O pad usage.
> >
> > 3. There is no reason why we only specify 10GE, which bonds 4 individual
> > links together. Probably there isn't any information source or sink today
> > which has to be specified at 10Gb/s. Why we have to glue 4
> > individual links
> > together just for making more problems in design, test, and maintain? The
> > only clue is the historical reason that Ethernet had migrated
> > from 10ME to
> > 100ME, and from 100ME to 1GE. Does only decimal system work for data
> > rate specification?
> >
> >  4. If 4 individual links at 2.5Gb/s are glued together to form a 10GE,
> > we need word synchronization, as someone suggested using Hari coding.
> > Transmission at 10Gb/s (or 12.5Gb/s with 8B10B coding) are more difficult
> > than sending 4 links individually. From reliability point of view, an
> > encapsulation of several links will greatly reduce network reliability.
> > For example, if one of the link is broken the whole frame of data is
> > destroyed, therefore, nothing could pass even if the rest three links
> > function well.
> >
> > 5. At today's technology, few devices could support 10GE within one link
> > for a meaningful distance. From technology point of view, there isn't any
> > remarkable breakthrough so that we can utilize a new and cheap device
> > to implement 10GE links. For most of the transmission media, what we
> > can utilize are the lower speed links. 10Gb/s data can only be
> > transmitted
> > at distance about a few dozens cm in PCB.  Long distance optical
> > fiber link
> > is an exception, and that means we only need 10GE at switcher interface
> > to the long distance trunks.
> >
> > 6. In fact, we could simply extend the Gigabit Ethernet standard to
> > 2.5/5 GE specification by adapting new speed as a smooth move. If we
> > use 8B10B encoding, the speed will be 3.125/6.25 Gb/s, respectively.
> > Simple and efficient! Why not?
> >
> > Thank you for your attention
> > Regards
> >
> > ___________________________________________________
> > Fenghao Mu
> >
> > e-mail: fenghao.mu@xxxxxxxxxxxxxx
> > SwitchCore AB         phone: +46 46 270 15 85(o)
> > Scheelevägen 32               fax: +46 46 270 15 81
> > S-223 63  LUND
> > SWEDEN                        homepage: www.switchcore.se
> >
> >