e1000 still

[Ed Brown]

On Sat, 2005-03-19 at 08:07, S. W. Davison wrote:
> We filed a support request with Red Hat.  Their
> response was basically that anaconda and the rest of
> the installer were working as designed, and that we
> should alter the parameters on our HP Procurve
> switches (turning "spanning tree" off and turning
> "portfast" on were suggested).

It's hard to understand this position by RedHat.  Anaconda is clearly
NOT working as designed.  This has been going on for more than a year,
with problems reported for ftp, http, and NFS installs, using HP, 3com,
and Cisco switches.  Altering switch configurations just to be able to
install is not always possible, much less a desirable solution.

The issue does seem to involve some subtle timing problem in anaconda. 
Efforts to capture traffic to pinpoint the breakdown fail because just
inserting a supposedly passive analyzer between the box being built and
the switch make the problem unreproducible.

As an aside, this problem has other consequences in an organization. 
Promoting wider use of RedHat among co-workers is a harder sell when
something like this goes on unresolved, and apparently unacknowledged,
for so long. 

Ed Brown, RHCE
Los Alamos National Laboratory