RE: [Gumstix-users] Duplicate MAC address problems

[J.P. N. <jn...@pr...>]

For now I will just address the multiple MAC problem, since that's
really what we're scrambling to fix.
=20
Out of about 50 units in the field trial, there are three or four pairs
of devices that share a common MAC, two triples (3 device with the same
MAC), and one quadruple.  In other words, there is MAC duplication on 18
out of 50 devices, which suggests obviously that there's a bug
somewhere.  There's a strong possibility that the bug is from our
software, but we need to solve the problem nonetheless.
=20
We're using U-boot 1.1.2.  Is this a problem?  Alternatively, if I
change the u-boot MAC, will that cause the Linux MAC to change?  I know
that many of the devices seem to share u-boot MACs.  The follow up
question is: is there a way tp change the u-boot MAC?
=20
Thanks,
JP Norair

________________________________

From: gum...@li...
[mailto:gum...@li...] On Behalf Of Craig
Hughes
Sent: Monday, April 03, 2006 01:43
To: gum...@li...
Subject: Re: [Gumstix-users] Duplicate MAC address problems


On Apr 2, 2006, at 5:54 PM, J.P. Norair wrote:

	Alright. . .  I'm aware that during boot a unit ID number is
transformed into the MAC address, but we have the problem that several
gumstix are taking the same MAC.  What is a good way to fix this?

Hmm.  The way the conversion is done, it should be really really rare
for collisions to happen, if you're using the latest u-boot code -- the
generated MAC should have really quite good randomness for all but 8
bits -- ie there are basically 40 bits of very random data; at least
assuming that my rotating-XOR thing is as clever as I think it is.  The
Taylor series approximation says that if you've got 10,000 gumstix, the
odds of a collision are approx. 1-e^-((10000*9999)/2*2^40) =3D 0.005%


Of course, that assumes that the MACs really are distributed uniformly
across the 40-bit space, which perhaps they're not.


	As an aside, the SMC91C111 chip, or at least some implementation
of the ethernet interface, has given us loads of trouble: now the MAC
addresses, before, a 1 in 20 chance of not being able to access the PHY.

What error were you seeing when not able to access the PHY?  What
voltage are you powering the netstix with?  Is anything getting really
hot (not just warm, but untouchably hot) on the board?


C