e1000 still

[Jeroen Roodhart]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

|I don't believe there is a need to wait 30 or 45 seconds for spanning
|tree calculations on the switch.  There is something much more subtle
|going on here.  As I mentioned before, we've tried capturing traffic
|with a passive analyzer between the box being built and the switch, and
|just this change to the network connections allow the install to >proceed
|normally (as does inserting a small 10/100 hub inline, or replacing the
|gig switch with a 100Mb switch.)  It isn't about spanning tree
|calculations on the switch, I think there is some sub-second change in
|the characteristics of the negotiation or anaconda's network setup, >when
|spanning tree is on or off, that make a difference.
|
|-Ed

I just want to support Ed here, this is the _exact_ behaviour that we
experienced.

I really don't see the issue here. I you retry try when it fails, you
just make the proces more fault tolerant (normal programming I think,
especially [as said before] when you use UDP).

If it works in one go, you don't have any adverse effects (it's just as
fast) and if it doesn't, it works with a short delay. What's the
deal?

In the mean time, when we experience this behaviour again (seems to
have dissapeared when going to FC3) I'll patch the code myself.

With kind regards,

Jeroen
- --
Jeroen Roodhart              University of Amsterdam
jeroen at science.uva.nl        Faculty of Science / ICT-Group
Systeem- en netwerkbeheer    Tel. 020 525 7203 / 06 51338165
- --
See http://www.science.uva.nl/~jeroen for openPGP public key
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.6 (GNU/Linux)

iD8DBQFCQToy37AP1zFtDU0RAtnwAJ9RipKbZxzWWSRsbyVW/I39B7sH6wCeN2BS
VAu/Rj9aha/xl0m37IcUGSE=
=TY1P
-----END PGP SIGNATURE-----