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

Re: [802.3_100GNGOPTX] 802.3bj Auto-negotiation consensus building



Hey Chris,

No disagreement on your first point. Arthur and I discussed the same thing offline. Rate switching is more prevalent in the electrical domain.

On the second point, the FEC used in 802.3ap has the same Ethernet MAC rate and the same lane rate as no FEC, but the latency impact is huge. Granted, there are many ways to do FEC and there is also the ability to use up IPG or manipulate the encoded data to keep the FEC and non-FEC line rate the same.

Most of this discussion is probably better saved for the task force. Maybe all that is needed at this point in time is to know that methods to switch between FEC on/off, if deemed necessary, can be developed.

Cheers,
Brad


-----Original Message-----
From: Chris Cole [mailto:chris.cole@xxxxxxxxxxx] 
Sent: Tuesday, December 06, 2011 12:14 PM
To: Booth, Brad; STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx
Subject: RE: [802.3_100GNGOPTX] 802.3bj Auto-negotiation consensus building

Hi Brad,

I have not seen the requirements for AN switching between 40GE and 100GE. It was pointed out that FC does AN switching to support two lower rates however that is a different application. We may consider first defining the Ethernet problem that needs solving, preferably in a presentation.

The AN switching requirement that arises with FEC is for the same Ethernet MAC rate but different lane rates, one without FEC and one with FEC, and only if the FEC rate is >0% overhead.

Chris


-----Original Message-----
From: Brad Booth [mailto:Brad_Booth@xxxxxxxx] 
Sent: Tuesday, December 06, 2011 8:14 AM
To: STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_100GNGOPTX] 802.3bj Auto-negotiation consensus building

Arthur,

It would depend on the number of speeds to negotiate. In 802.3ap, it was for the single lane 1G and 10G. In this effort, there are four lanes. There is only one four-lane optical PMD (LX4) that operates at 10G. The others are 40G (4x10) or 100G (4x25). IMHO, support for 10GBASE-LX4 is very, very low priority; therefore, the speed selection is only between 10G and 25G.

Clause 37 has too much baggage, and modifying Clause 73 for optics seems to be overkill.

Cheers,
Brad



-----Original Message-----
From: Arthur Marris [arthurm@xxxxxxxxxxx<mailto:arthurm@xxxxxxxxxxx>]
Sent: Tuesday, December 06, 2011 05:33 AM Central Standard Time
To: STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_100GNGOPTX] 802.3bj Auto-negotiation consensus building

Brad,
   Thanks for the clarification. I think the parallel detect specified in 802.3ap would be inadequate for optics speed negotiation. The 802.3ap parallel detect function was only intended to detect two legacy PHY types that could interoperate with the newly defined 802.3ap PHYs. I don’t think it would scale to support a large number of PHY types and it does not allow a PHY to advertize more than one speed of operation. Clause 8 of the Fibre Channel Framing and Signaling specification describes a method of link speed negotiation where the transmitter shows off its abilities by operating at different speeds in sequence while the receiver detects the one it wants to use, however this seems fairly complex and I think it would make more sense for 802.3 to reuse Clause 37 for optics speed negotiation and to contradict what I said below to consider if the Clause 73 AN protocol could be used over optics.

Arthur.

From: Brad Booth [mailto:Brad_Booth@xxxxxxxx]
Sent: 05 December 2011 15:40
To: STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_100GNGOPTX] 802.3bj Auto-negotiation consensus building

Arthur,

Clarification:

-          I support using Clause 73 AN for P802.3bj.

-          Sequence ordered sets could provide fast FEC on/off switching that having to restart AN.

-          Sequence ordered sets could provide a means to exchange FEC abilities, etc. in optical media.

-          Speed negotiation in optics could use parallel detect specified by 802.3ap.

Hopefully that clarifies things a bit better.

Thanks,
Brad


From: Arthur Marris [mailto:arthurm@xxxxxxxxxxx]
Sent: Monday, December 05, 2011 5:44 AM
To: STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx
Cc: Booth, Brad
Subject: RE: [802.3_100GNGOPTX] 802.3bj Auto-negotiation consensus building

Brad,
   Let me make a few comments.

   For 802.3bj it is necessary to use Clause 73 for auto-negotiation because Clause 73 already supports port types for four-lane twinax and back-plane. If new 100G port types are being added for these media, they need to be included in Clause 73. Clause 73 also gives you FEC negotiation for free.

   Clause 73 clearly does not work over optical media.

   You can use ordered sets for exchanging ability and requests, but this only works if both sides of the link are running at the same speed and the error ratio is not too bad. For next generation optics you are going to need to decide whether you need to negotiate speed in addition to determining FEC operation (for example you might want to negotiate between 40GBASE-SR4 and 100GBASE-SR4). If you do decide to negotiate speed you could consider re-using Clause 37 for auto-negotiation (which operates at a 1.25G data-rate).

   If you are not negotiating speed but are just concerned about bit errors, then the task being performed is really link training rather than ability resolution.

Arthur.


From: Brad Booth [mailto:Brad_Booth@xxxxxxxx]
Sent: 02 December 2011 19:25
To: STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_100GNGOPTX] 802.3bj Auto-negotiation consensus building

Stephen,

It depends on the FEC protocol adopted, implementation requirement, and the associated latency. The discussion got started around the idea of using Clause 73 AN for FEC ability exchange in the optical domain.

Considering we’re still in the study group phase, some of those questions cannot be answered as we have not adopted an FEC proposal. If an exchange protocol is required, then IMHO the use of sequence ordered sets would be preferred over creating of a new autoneg protocol.

Cheers,
Brad