Re: [802.3_100GCU] EEE work in IEEE P802.3bj
Thanks Mark!
-----Original Message-----
From: Mark Gustlin (mgustlin) [mailto:mgustlin@xxxxxxxxx]
Sent: Wednesday, October 19, 2011 3:03 PM
To: Booth, Brad
Cc: STDS-802-3-100GCU@xxxxxxxxxxxxxxxxx
Subject: RE: [802.3_100GCU] EEE work in IEEE P802.3bj
Brad,
The stated purpose of refresh from az is:
While the PHY is in the LPI mode, the PHY periodically transmits a refresh signal to allow the remote PHY to refresh its receiver state (e.g., timing recovery, adaptive filter coefficients) and thereby track long-term variation in the timing of the link or the underlying channel characteristics.
I can't think of a reason for why the receiver needs to worry about achieving alignment during a refresh cycle, so I don't think it is necessary to send Alignment Markers in a short or normal interval. So we have flexibility to do whatever we think is easiest during the refresh periods.
Thanks, Mark
From: Brad Booth [mailto:Brad_Booth@xxxxxxxx]
Sent: Wednesday, October 19, 2011 11:46 AM
To: STDS-802-3-100GCU@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_100GCU] EEE work in IEEE P802.3bj
Mark,
Unfortunately, I'm going to be unable to make the call tomorrow.
I like the general concept you have developed. Do you have any information relative to how much drift there would be on the alignment marker going in and out of LPI?
Just trying to figure out if the Rapid AMs provide any value in the refresh cycles.
Thanks,
Brad
From: Mark Gustlin (mgustlin) [mailto:mgustlin@xxxxxxxxx]
Sent: Wednesday, October 19, 2011 1:05 PM
To: STDS-802-3-100GCU@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_100GCU] EEE work in IEEE P802.3bj
All,
My slides for tomorrow's call are attached.
Thanks, Mark
From: Michael Bennett [mailto:mjbennett@xxxxxxx]
Sent: Tuesday, October 18, 2011 12:23 PM
To: STDS-802-3-100GCU@xxxxxxxxxxxxxxxxx
Subject: [802.3_100GCU] EEE work in IEEE P802.3bj
Folks,
The purpose of this note is to inform you of the work that has been done so far regarding EEE for IEEE P802.3bj. A group if individuals have been working on various ways to accomplish EEE for this project. I have included below the notes from the last conference call we had and the presentation material we discussed. There will be another call on Thursday, October 20th at 8 AM Pacific Time. It is scheduled for 1.5 hours. Please contact Hugh Barrass or Mark Gustlin for dial-in information. Mark Gustiln will review his work and take feedback from the group. He will make his slides available sometime between now and the call. We will have another call before October 28th and will run a doodle poll to select the best day and time for that call. I will be unavailable indefinitely as I am dealing with a family emergency. Please contact Mark for your questions regarding the call. Questions regarding the work should be directed to the reflector.
Best regards,
Mike
------------------
Here are my notes from the call last week (9/28/11, approximately 10:05- 11:25 PST):
----------
On the call: Adam Healey, Arthur Marris, Brad Booth, Hugh Barrass, Matt Brown, Pedro Riviriego, Wael Diab, Mark Gustlin and Mike Bennett
Thanks to Hugh and Cisco for hosting the call.
Pedro presented "Energy Efficient 100G with Modular LPI"
Comments and Questions:
General discussion on latency. As latency increases tolerance decreases. There is a trade-off between latency and power savings
Some disagreement whether or not 3 lanes could be active (only one lane turned off)
Need to prove the assumption that the multi-lane approach would lower the transition time
Question as to whether or not there are any savings to to powering off FEXT cancellers.
We need more data center traffic for analysis. Traffic data is hard to get. Pedro has a script that will take a capture as input and output a text file with frame inter-arrival times ad frame sizes. Removing everything else may alleviate concerns about sharing the data.
Question - how fast is the LPI transition? What is the minimum wake time? fair amount of discussion but no definitive answer.
One thing many on the call agreed on was the need to avoid complexity. Don't want a new PCS. Some things may be able to be done in the PMA instead.
Discussed whether or not we should ask for an objective, and the advantages and disadvantages of doing so.
Discussed the use of IEEE P802.3bj Task Force Ad-Hoc in order to make sure discussion and material are available to anyone.
------------
Post meeting:
Next Steps:
Get data center traffic for analysis
Review Mark Gustlin's LPI presentation
Continue to build consensus for EEE for 100G Backplane and Twinaxial Copper Cable
Next Call - 10/20
Doodle poll for call before 10/28/2011
--
Michael J. Bennett
LBLnet Services Group
Lawrence Berkeley National Laboratory
Phone: 510.486.7913