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

On Tue, 8 Jul 2003, Ted Lemon wrote:

> 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.

[snip]

> 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.
>

There may already be a way around this.

> 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.

Agreed.

The problem here is two fold:

(1) xids cannot be guaranteed not to collide because some clients
    cannot be programmed to generate enough randomness.

(2) Some datalinks, like firewire, do not allow setting chaddr to
    anything reasonable.

We already have the DHCP Hostname option. Clients can be
configured to pass this option when the admnistrator expects high
xid collisions. The DHCP server can also be configured to assign
leases per the DHCP Hostname option. We just need to promote the
use of the DHCP Hostname option to be used to counter this
problem.

Is this too kludgy?

Barring thing it would make more sense to add a DHCP option for a
larger unique ID (Secondary XID?) than to modify the BOOTP
header.

-- 
Thamer Al-Harbash

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