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

Re: Unified PMD vs. Unified PHY

>The original compromise was to have separate LAN and WAN PHYs.  Now a group
>of people is trying to compromise on the compromise by compromising the best
>features of both of the separate PHYs in a continuing attempt to justify an
>8B10B precoding to everything else by calling it another name.  This is the
>same problem that was had with the "hari" proposals.  The "UniPHY" as
>proposed is nothing more than the original 10.00 Gb 8B10B LAN PHY attempting
>to "shoe horn" itself into the WAN PHY and being an additional 3% less
>efficient doing it.

A group of people is trying to find a solution that meets all of the
objectives that were established by the HSSG, and they are trying to
provide a platform on which concensus can be established.  The
proponents of the UniPHY have worked hard to understand and balance all
of the requirements and realities.

UniPHY does not require or depend upon 8B/10B encoding, in any way,
shape, or form.  In the presentations regarding UniPHY, we have shown
the XAUI (and associated 8B/10B encode/decode) as an option.  Nothing 
prevents the UniPHY from being connected directly to the XGMII, if that 
is what an implementer wants to do.  In that configuration, 8B/10B would 
not be used, at all.

The 3% cost in efficiency relative to the length/type/hec proposal has
nothing to do with 8B/10B encoding either.  It stems from a desire to
employ a robust frame delimiting mechanism that does not require
substantial changes to the MAC or substantial changes to the Ethernet
frame format.

I see it this way: UniPHY meets all of the objectives we established
for the WAN PHY, and has all of the advantages of the length/type/hec
delimiting proposal in terms of compatibility with the photonic 
infrastructure, while providing 97% of the throughput.  97% is pretty
darn good, considering the other advantages afforded by UniPHY.

>I know that you have an emotional attachment to 8B10B.  I know that at least
>one large LAN vendor has already done an ASIC using 8B10B out of the MAC.
>Not using 8B10B does not preclude the LAN vendor from having his LAN PHY
>based on block encoding, 8B10B or 64B/66B; while allowing the WAN vendors
>from having a non-8B10B solution.  Please do not presume that one solution
>will be best for all implementations.

I would love to know who this mystery vendor is. They have been mentioned
a couple of times.

Howard Frazier
Cisco Systems, Inc.