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

Re: [802.3af] Link negotiation

From:  Scott Burton@MITEL on 12/11/2001 05:24 PM
Hi Mike,
We've done some link timing tests in-house with combinations of PHYs from three
different vendors over 2m and 90m of cable. The link up time was taken to be
that from when the "PD" PHY power rail first appeared to when the Link LED on
the "PSE" was activated.

Under these conditions we found that the time to link for the various
combinations varied from 1.7 seconds to 5 seconds, with the fastest time being
between two PHY demo boards. We also found instances during link up where the
link would appear, then drop for a second or so, then reappear, but its possible
this may have been due to a software incompatibility with a specific PHY.

I'm not sure if these times are representative of what can be expected in
practice, but if so then it suggests that using link as a disconnect detection
method may be somewhat problematic.

Mike_S_McCormack@xxxxxxxxxxx on 11/26/2001 10:47:32 AM

To:   stds-802-3-pwrviamdi@xxxxxxxx
cc:    (bcc: Scott Burton/Kan/Mitel)

Subject:  [802.3af] Link negotiation

At the last plenary, we decided to allow PSE's to evaluate the "PD_DATA_LINK"
variable as a qualified for the power request signature.  We added the
requirement without modifying any of the timing parameters and at least need to
provide a hold off time from power assertion to "link up" since we need to
recognize that PDs can't establish link without first getting power.

On return to my office I've checked with a few people more knowledgeable than
myself and have been told that link negotiation, on a proper arrangement of
cable and equipment, of 600ms has been observed in the lab (100BASE-T.)  This
time to establish "link up" does not include such real world problems such as
plugging two 10/100/1000 devices together via a 100m of CAT-3 cabling and
letting things go their course.  I will take it upon myself to try to dig timer
values out of the 802.3 specification so that we can bound what we are already
committed to, but I'd like to hear form others with real world experience or
knowledge of what their PHYs were designed to.

There are other real world considerations, such as OSs which routinely reset
their MAC chips after the BIOS has already enabled them and cause "link" to drop
out.  It would also appear the my Windows98 PC resets the MAC when I run
Winipcfg and ask it to release and renew all (it looks like the link LED
flickers, I need to check with an analyzer.)  These types of behavior would
cause units to get powered down in the middle of otherwise normal operations.

I'd suggest we scrutinize what we have really done when we say that PDs must
maintain the "PD_DATA_LINK" variable.