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

Re: Hari and train-up sequences




Linda, Brad,

I agree with Linda's direction to treat a 10 GbE link consisting of two Hari
interfaces interconnected by a PMD medium based link as a single link and to
define a link initialization protocol at the PCS layer. I don't see a reason to
separately initialize the individual Hari and PMD medium interfaces.

Initialization sequences in Gigabit Ethernet, based on Fibre Channel, which is in
turn based on IBM's ESCON are very similar in nature and are essentially the same
as that proposed by Linda in her last note (quoted below). This sequence is
essentially implemented by the Auto-Negotiation protocol specified in 802.3z
clause 37. For a trip down memory lane, take a look at the Link Startup Procedure
illustrated on page 9 of PCS proposal for Gigabit Ethernet:
http://grouper.ieee.org/groups/802/3/z/public/presentations/nov1996/RTpcs8b_sum5.pdf

It may look complex, but one can't argue with the success of these standards. I am
very hesitant to endorse a much more complex standard (10 GbE) which does not
provide the basic ability to reliably recover from common link errors.

Separately, note that AN information exchange is actually data transported over
the Link Initialization "carrier" and should be considered on its own merit. In
the past, three classes of information have been transported using various
Auto-Negotiation protocols:

1) Link speed, protocol, and configuration determination which, if required, must
be exchanged before link level protocols can execute;
2) Link faults require the exchange of a least link protocols;
3) General statistics, performance information, vendor IDs, etc. do not require
link protocols and can be exchanged via Ethernet packets.

Careful though needs to be given as to the requirements for the exchange of
traditional AN information.

Best regards,
Rich

--

Linda Cheng wrote

> Hi Brad,
>
> I'd like to see a link initialization defined at the PCS layer
> between a local device and link partner.
>
> It's something we need and something which doesn't have to
> be complex.
>
> A two-way handshake may look like 1) one end sends a stream of Link
> messages indicating it is in the init state 2) the remote side must
> respond with the Link message (some 10b codegroups chosen for this) and
> after some time Link messages with acknowledges 3) when the local side
> receives these it sends Link acknowledges for some time 4) when
> either side receives Link acknowleges they are done and send Idles.
>
> So the PMA would need to be ready before the Link message with
> acknowleges are sent. By HARI I assume you mean PMA. HARI is the
> interface between the PMA and PMD.
>
> Linda
>
> > From: "Booth, Bradley" <bradley.booth@xxxxxxxxx>
> > To: stds-802-3-hssg@xxxxxxxx
> > Subject: RE: Hari and train-up sequences
> > Date: Tue, 16 Nov 1999 13:01:03 -0800
> >
> > I need some clarification here.  Are you proposing a link initialization for
> > the Hari?  If so, are you also proposing a link initialization for the local
> > device to link partner connection?  If you're proposing both, then are you
> > proposing to have 3 stages of link initialization?  If so, I believe that
> > this may add more complexity than we need to deal with.
> >
> > Cheers,
> > Brad

> ---------------------------------------------------------------
> Linda Cheng                     Cisco Systems
>                                 Desktop Switching Business Unit
> (408) 527-2015 (phone)          170 West Tasman Drive
> (408) 527-4698 (fax)            San Jose, CA 95134-1706
> ---------------------------------------------------------------

  ----------------------------------------------------------

Richard Taborek Sr.   1441 Walnut Dr.   Campbell, CA 95008 USA
Tel: 408-370-9233     Cell: 408-832-3957     Fax: 408-374-3645
Email: rtaborek@xxxxxxxxxxxxx