[syslinux] Network Boot IP Configuration Dilemma

[H. Peter Anvin]

Fred Feirtag wrote:
> 
> In fact, extraneous leasings are the sole source of the actual
> problems I've seen.  A system that dishes out IP leases occasionally
> becomes active on our network, and it occasionally answers
> my booting clients before my static server.  My focus has been
> how to identify and respond to this problem.  The error report
> will never be "someone plugged a leasing IP server into the
> network" it will be "my workstation is hanging sometimes."
> Reducing the numer of ways a workstation hangs on bootup
> is a good thing for anyone having to diagnose trouble calls.
> 

The thing is, if someone is running a rogue DHCP server on your network,
you'd have people's workstations hanging anyway.

> I would rather create 10 potential, but solid, hardware configuration
> errors that, once fixed stay fixed, than create one intermittant
> operational error, or even as in this case, make one intermittant operational error
> harder to diagnose!  I freely admit that is a strong bias, but that's the
> "Dilemma" of my original subject.  That is what I see at issue between
> the single vs. double dhcp request, by depending on identical duplicate
> dhcp responses.

You're not depending on it any more than you are depending on proper
DHCP operation in general.

> From what I can tell, there are good reasons to do it either way.
> (The argument by James that the kernel might understand and
> potentially utilize more items than PXE is a good one, I think.)
> If IPAPPEND is going to be provided as a feature, it shouldn't
> be crippled.  I really, really don't see providing for passing useful
> parameter values as "plugging random crud" into it.

Perhaps I should explain that the only reason IPAPPEND is there is to
support a small set of users with a particularly screwed up network
infrastructure, which uses a dedicated IP address pool for "PXEClient"
install clients.  It's a dirty hack and if you have to use it, you have
a serious problem to begin with.  Given the number and the kinds of
support requests it has generated, I'm getting pretty convinced that it
was a bad idea.

	-hpa