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

Re: [802.3BA] Discussion on 40G for => 10 km SMF

Hi Dan,
I think it is only a matter of the precise ratios of FITs, not any
substantive difference in the operational procedures.

In the case of electronics failures, in either case there are probably
some failure modes that can still send light down the fiber and others
that can't for serial or 4x10G CWDM.

In the case of optics failures, nearly all failure modes on a serial
transmitter would not send light down the fiber, while many optics
failures in CWDM could still send light down the fiber (perhaps an
advantage to CWDM in distinguishing fiber failures from interface

But for both, there are a set of failure modes that can still send light
down the fiber and a set of failure modes that cannot. Perhaps a higher
percentage of failures on the serial interface cannot.

Even seeing light on the fiber doesn't necessarily mean the glass is
good. You might still get some light on a fiber path that is impaired
due to dirty connectors, excessive bending, etc. that doesn't give you
good enough OSNR to recover the signal.

-----Original Message-----
From: Dove, Daniel [mailto:dan.dove@xxxxxx] 
Sent: Friday, August 08, 2008 1:42 PM
To: Trowbridge, Stephen J (Steve); STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
Subject: RE: [802.3BA] Discussion on 40G for => 10 km SMF

From a diagnostic perspective, having light on three channels would be
more useful than having none on the serial link.

In the former case, I know the glass is good and most likely it's a
transceiver that has failed. In the latter case, I cannot discern the

-----Original Message-----
From: Trowbridge, Stephen J (Steve)
Sent: Friday, August 08, 2008 12:06 PM
To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3BA] Discussion on 40G for => 10 km SMF

Hi Mikael,
Operationally, we have a simplex interface. There is no internal
redundancy, no logic that lets you operate in any kind of reduced
bandwidth "limping" mode.  There is no per-wavelength replacable unit.
An optical failure is no different from an electronics failure. If you
lose one laser, you are just as dead as if you have a stuck bit on one
of your four virtual lanes (which could happen just as easily if the
interface were serial) and you can't frame all of the virtual lanes and
put the signal back together. You don't need to even know which of four
lasers failed, because the repair action is the same in any case.

The serial technology being proposed does not extend to colored
interfaces in the same way we saw at 1G and 10G. In the first place, it
is 1310nm and for DWDM you would need 1550nm. At 1550nm, you couldn't
get the reach you want for 40G without dispersion compensation (at
1310nm you are at the zero dispersion point in the fiber). The EML and
NRZ modulation format are not what you would want at 40G because
customers want 50GHz spacing between wavelengths. The technology you
would want for a DWDM solution shares very little with what is being
proposed for 40GbE serial, and choosing serial would do almost nothing
to reduce the cost of that application.

So I don't think these are the aspects that should drive the decision.
The issues are cost and time to market, and on that basis, I think the
clear choice is 4x10 CWDM.

-----Original Message-----
From: Mikael Abrahamsson [mailto:swmike@xxxxxxxxx]
Sent: Friday, August 08, 2008 12:03 PM
To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3BA] Discussion on 40G for => 10 km SMF

On Fri, 8 Aug 2008, Jesse Simsarian wrote:

> 1. Operationally, 4 x 10G and 40G serial are equivalent.  If anyone is

> voting for serial because of operational concerns, please bring this 
> out on the reflector.

I have two questions (trying to determine operational impact for me as

In a 4x10G module, if one laser or receiver breaks, what happens then?
Will the port shut down the same way it would with a 40G serial?

Also, future adhoc standard for 40G serial CWDM optics (ie someone
developing 40G serial CWDM-band colored optics the same way that has
happened on 1G and 10GE) speaks in favour of serial instead of 4x10G, or
am I mistaken?

Mikael Abrahamsson    email: swmike@xxxxxxxxx