[802.3EEESG] Need for models to have accuracy in transition time estimates
Unfortunately Aquantia is not able to participate in the Atlanta
meeting. But there are a point we'd like to pose for discussion on the
subject of transition time.
The topic of this post is in understanding the absolute and relative
transition times between Subset PHY and Rapid PHY Selection (RPS). For
both schemes, the transition time from GE to 10GE depends on how much
re-adaptation needs to be done to get back the optimal settings of
filters in 10G mode.
At the first order, the transition time will depend on two major
factors: the changes in channel characteristics and the changes in
noise sources.
The primary advantage of the subset PHY scheme is that it should be able
to be designed in a manner that permits it to track changes in the
channel. RPS between GE and 10GE does not have this advantage.
However, neither scheme can anticipate the changes in the relative PSDs
of signal and noise sources that must be adapted to at the time of the
switch. This point was not properly considered in earlier
presentations, including Aquantia's, which projected estimates of
potential transition times in the range of microseconds. These can
potentially be much more than a millisecond, depending upon the
definition of a change.
The primary difference in transition time between Subset PHY and RPS
(and the primary advantage of Subset PHY) then appears to be the ability
to track the channel. We believe channel related changes will generate
longer times than noise related changes.
There are two points we need to develop further, both dependant on
developing better models for the environment we are addressing.
a) Model for change in the channel. What defines the maximum
permissible change in the channel which must be accommodated? What
physical changes? What temperature changes? Of course length will not
change, and fixed parameters (PBO, THP) are static.
b) Model for change in noise sources. Changes in transmit power,
residual echo, NEXT and FEXT. Additional AFE noise, nonlinearities and
jitter in full-duplex mode. What alien noise source changes?
A subset PHY mode may have a greater impact on OEMs and systems users
(qualification and accommodating new modes), and needs to be justified
by delivering a clear benefit over RPS (which utilizes existing PHYs).
Discussing any transition time benefit quickly implies the use of an
agreed-to model, or a common definition which permits development of
meaningful estimates.
With these definitions, it is possible to estimate both the absolute and
relative transition times. If they are dominated by changes in the
channel, then the subset PHY may be preferable. If it is dominated by
changes in noise sources, then the subset PHY may not offer sufficient
advantage and RPS may be a better approach.
Regards, Bill
Bill Woodruff, (c) 408 582-2311
Aquantia - VP Marketing, 408 228-8300 x202