Re: portfast

From: Mikael Abrahamsson (swmike@swm.pp.se)
Date: Sat Jul 07 2001 - 05:57:34 EDT


On Sat, 7 Jul 2001, Ryan O'Connell wrote:

> On Sat, Jul 07, 2001 at 11:23:33AM +0200, Mikael Abrahamsson wrote:
> > I thought this was accounted for in the DHCP specs, that you need to wait
> > at least 30 seconds before deciding that the dhcp server is unavailable. A
> > lot of network cards do not do "link-up" until the driver initialises the
> > card and thus you're in the 30 second wait period when you bring up the
> > link and then starts asking for dhcp servers.
>
> IIRC, Spanning Tree takes 30 seconds to go into forwarding state in the absence

Yes, that is the default setting (probably in the RFC).

> Some Operating Systems (Windows NT 4 seems to be one, although I suspect it was
> the fault of the driver, not the OS Kernel) actually start the DHCP process
> before the network card has fully initialised, so you get less than 30 seconds.
> I've actually seen one extreme case (With 3Com 3C905s - both the B and the C)
> where even the speed/duplex auto-negotiation phase is enough delay to cause
> Windows to fail booting. It's an extreme case but it does happen. (DHCP
> eventually suceeded, but the PDC/BDC search phase had already timed out)

Then I believe those implementations are broken. Spanning tree has been
around forever and this forwarding delay should be known by all
and implementors should account for them.

I checked the DHCP/BOOTP RFCs but could find no values that
implementations are required to use when it comes to timeouts etc. I find
this odd. Making it a SHOULD or perhaps even a MUST to have at least 60
seconds timeout would probably be a good thing.

-- 
Mikael Abrahamsson    email: swmike@swm.pp.se



This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:44 EDT