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

Re: Hari




Rich:

This looks like a reasonable design based on the architecture I've
outlined. The architecture provides a base for understanding all the issues
surfacing in email about Hari and provides insite on what is required to
use Hari for general purpose applications. I still believe Hari is a PHY
over backplane copper not layer interface.

Cheers,

Paul

At 04:23 PM 11/23/99 -0800, Rich Taborek wrote:
>
>Paul,
>
>Boy this is taking a lot of cycles out of my MAS budget :-(
>
>I'm still very confused about your notion of a Translator/Bridge/Relay
required for Hari implementations. Let me try another illustration by
comparing 10 GbE Hari to the serial output of the GbE Ten Bit Interface
(TBI). Note that only the transmit direction is shown for simplicity.
>
>
>1 GbE, PMA (SerDes) separate from PCS
>-------------------------------------
>
>                         TBI
>+---------+          +---------+           +---------+
>|         +---------->         | 1 serial  |         |  Medium
>| GbE PCS | 10 lines | GbE PMA +-----------> GbE PMD +---------->
>|         +---------->         |   8B/10B  |         |
>+---------+ 125 MHz  +---------+ 1.25 Gbps +---------+
>
>
>1 GbE, PMA integrated with PCS
>-------------------------------------
>
>+---------+           +---------+
>|         | 1 serial  |         |  Medium
>| GbE PCS +-----------> GbE PMD +---------->
>|  & PMA  |  8B/10B   |         | 1 serial
>+---------+ 1.25 Gbps +---------+
>
>
>10 GbE, PMA integrated with PCS
>-------------------------------
>
>+---------+   Hari   +---------+
>|         +---------->         |   Medium
>| 10 GbE  +----------> 10 GbE  +------------>
>| PCS/PMA +---------->   PMD   | 1-4 serial
>|         +---------->         |
>+---------+ 4 serial +---------+
>             8B/10B
>           3.125 Gbps
>
>Several 10 GbE PMDs, like GbE support the propagation of the 8B/10B coding
over the medium. These may include Parallel Optics, WWDM and 12.5 Gbps
Serial. However, those are PMDs and Hari is an interface. Other 10 GbE PMDs
strip off the 8B/10B code in the PMD and recode with more efficient codes.
These include MAS, ~10 Gbps Serial LAN PHY and the WAN PHY.
>
>I beg to differ about "corrupting" Ethernet layering to fit Hari in my
previous note. No additional layers are added. I've simply added an
optional coding and signaling function to the PMD sublayer and illustrated
the optional serial interface (Hari) between the PMA and PMD sublayers.
>
>The last point in your note argues that the 16 bit parallel interface is
very different architecturally from Hari. I have shown above that this is
clearly not the case. In addition, Hari has the distinct advantage of <1/4
the pins, drivers and receivers; longer trace lengths; no clock needed with
the data; built in error control; DC balance; and special character support.
>
>I have no problem with supporting the 16-bit parallel interface as an
additional optional interface between the PMA and PMD if you feel that it's
more appropriate for specific PMDs such as the WAN PHY.
>
>Best regards,
>Rich
>
>--
>
>Paul Bottorff wrote:
>
>> Rich:
>>
>> We agree that Hari has qualities about it that make it desirable for a
section of the market. We disagree that Hari is equivalent to a parallel
PMD interface. I believe both these interfaces have their place. They are
not mutually exclusive, they are complementary.
>>
>> As far as the architectural model, I'm just applying the 802.3 model
exactly, rather than corrupting the layering to fit Hari. To make the model
complete Hari can be viewed as performing peer dialog between 802.3 layers
as follows. This helps explain the existence of the L1 Translator (or
Bridge or Relay, choose your words) function which is currently required
for Hari implementations. The jitter elimination buffer is part of the L1
Translator while deskew is part of the Hari-PCS. I believe this gives a
much better understanding of how Hari behaves with respect to historic
Ethernets. The peer model is consistent with all the functions currently
presented for Hari. The model is as follows:
>>
>> 802.3 Peer Dialog
>>
>> ----------- ------------
>> | MAC | - - - - - - - - - - - - - - - - - - - - - - - | MAC |
>> |---------| ------------------------ |----------|
>> | Recon | | L1 Trans(Bridge) | | Recon |
>> |---------| |----------------------| |----------|
>> |Hari-PCS | - - - - - | Hari-PCS | Other-PCS | - - - - - |Other-PCS |
>> |---------| |----------|-----------| |----------|
>> |Hari-PMA | - - - - - | Hari-PMA | Other-PMA | - - - - - |Other-PMA |
>> |---------| |----------|-----------| |----------|
>> |Hari-PMD | - - - - - | Hari-PMD | Other-PMD | - - - - - |Other-PMD |
>> |---------| |----------|-----------| -----------|
>> |______________________| |_________________________|
>> 4 Wide Copper - 20" Fiber
>>
>> When this model is applied to a LAN PHY transform the Hari-PCS and
Other-PCS are identical and therefore require no transform. When applied to
a WAN PHY the Hari-PCS and the Other-PCS are different are require a
transform. The model works nicely in both cases. Only the implementation
changes.
>>
>> This model also makes it clear that the 16 bit parallel interface is
very different architecturally from Hari. Hari is a complete PHY of its own
while the 16 bit interface is a N-Layer boundary between PMA and PMD.
>>
>> Cheers,
>>
>> Paul
>
>  ----------------------------------------------------------
>
>Richard Taborek Sr.   1441 Walnut Dr.   Campbell, CA 95008 USA
>Tel: 408-330-0488 or 408-370-9233           Cell: 408-832-3957
>Email: rtaborek@xxxxxxxxxx or rtaborek@xxxxxxxxxxxxx
>
>
>
Paul A. Bottorff, Director Switching Architecture
Enterprise Solutions Technology Center
Nortel Networks, Inc.
4401 Great America Parkway
Santa Clara, CA 95052-8185
Tel: 408 495 3365 Fax: 408 495 1299 ESN: 265 3365
email: pbottorf@xxxxxxxxxxxxxxxxxx