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

Re: Hari


I am not necessarily excited about HARI either, but you are a touch away from reality on a few
of your points.

Roy Bynum wrote:

> Rich,
> I'm not sure what gave you the idea that Hari was favorable to serial PHYs.  I do know that
> it is not favorable to anything other than the legacy fiber channel 8B10B LAN PHY.  Hari is
> NOT a PHY neutral device interconnect.  Some LAN vendors have very huge boards and they want
> something that will support their 8B10B encoding over the distance of their board, and also
> the back plane if needed.  By introducing Hari as the standard for device interconnect
> between the PMA and the PMD,  they are specifically, and possibly knowingly, hampering the
> development of the agreed on WAN PHY.  There is no mechanism for rate controlling over
> Hari.  This is in violation of the agreed objective in this regard.  Hari is an attempt by a
> specific camp to control the development of 10GbE and limit development of anything other
> than the 10.0 only rate PHY.

Your focus is the operation of the phy, the features that go along with that, and your customer
base.  As always, most in your shoes forget about 'signal integrity'.  The HARI interconnect
allows the board designer to partition the board logic in an efficient manor to the ASICs and
FPGAs with minimal considerations to the geometry of the circuit board.  If I interconnect my
system at 10gig from point to point to allow the logic designer to use the 'perfect phy', I will
tripple the system cost .... and the marketing guys will tell you what that means to the end
cost - not my area.

I don't really like the interface, either, but you know what .... for the design of most systems
currently in or soon to be under development, the interconnect scheme seems to really fit well
and solve a lot of issues.  Is it perfect?  No, but I'll be damned if I have been able to come
up with something better given all the variables of the problem.

> As you will remember, I called the presenters and floor on the issue of what Hari really
> was.  It was admitted that it was a LAN extension interconnect by the individual that
> responded.  For those of us that are attempting to bring 802.3 into the 20th century by
> making a truly ubiquitous MAC, we have been astounded by the brazen push of this
> "interconnect".   Chip makers will confirm that Hari specifically makes the WAN PHY
> extremely difficult to implement.  Since Hari truly is not common to all of the PHYs as
> specified in the objectives, I suggest that it be withdrawn from consideration as part of
> the 10GbE standard.

I have not heard anyone say this, and I have been involved with several of them.  I have heard
individuals admit that they do not care for this interface, but each of them has also said they
can not think of anything better.

I would offer that we look at the pros and cons of what this interconnect can provide in our
designs, and then stand back and decide where to improve it.  And if you still don't like it,
well, I would ask you think about the signal integrity aspect of it for a while and not a
network manager ...... How would you like two or four card chassis that are inefficient in your
rack spaces?  What if the board size had to drop to 5in by 5in - the network processor won't be
what you would like to use .... The bottom line is this - 12.5gig is very easy to implement in
the lab with a controlled setup.  Production is very different.  The board sizes, the parts
used, the power sub-system, the connectors - all combine to impose some very impossible hurdles
to jump.  A 'HARI' scheme allows someone in SI to pick apart the problem into manageable

Joel Goergen