Re: [dhcwg] "client identifier" in server replies

On Tuesday 08 July 2003 07:49, Bud Millwood wrote:
> Filling out chaddr and htype *will* eliminate the problem. As long as
> client id is also present, the device will be identified with client id
> (which is the preferred identifier).

Bud, there are some cases, particularly with firewire, where there is no way 
to fill out the hardware address.   Of course it needs to be filled out in 
other cases - the protocol requires it.

> Barring that (and it shouldn't be barred), randomizing xid is not the only
> way to reduce clashes. Timing & relay segmentation further reduce the
> probability of a clash. And unicast responses, when received, are
> guaranteed to be for you.

It's not always easy to see that a response has been unicast.

In response to one of your earlier points, adding special cases to watch the 
wire doesn't help, because protocol specifications have to work properly in 
the general case, not just in special cases.

The solution of having the server notice xid clashes and stagger the replies 
unfortunately doesn't work because the first reply is going to be seen by 
both clients as a reply to its DHCPDISCOVER.

_______________________________________________
dhcwg mailing list
dhcwg@ietf.org
https://www1.ietf.org/mailman/listinfo/dhcwg