>
> MAC layer addresses are only valid on your local Ethernet segment.
Certainly, MAC addresses cannot be assumed valid beyond that point.
Some non-IP routing protocols, such as that used by XNS, actively
modify the MAC address. DECnet also does, for somewhat different
reasons. Both are attempts to avoid the need to assign unique
network layer addresses, but have become a kluge to watch for in
multiprotocol environments.
Let's assume a two-segment XNS network:
XNS Network #100
-----------------------------------------------
|
| Interface with burnt-in address AA-AA-AA-AA-AA-AA
-----
Router
------
|
| Interface with burnt-in address BB-BB-BB-BB-BB-BB
-----------------------------------------------
XNS Network #101
When the router initializes, it will copy the MAC address of the first
interface (let's call that AA-AA-AA-AA-AA-AA) to _all_ other interfaces
of the same router.
The network layer address of the first port becomes
100.AA-AA-AA-AA-AA-AA
and that of the second is
101.AA-AA-AA-AA-AA-AA
In principle, there will be no problem, because the "seed" MAC
address on the first interface is a unique burnt-in address.
If, however, a local administrator changed the MAC address on
the first interface before XNS came up, and that manually set
address was non-unique,all kinds of problems can ensue.
The DECnet algorithm is sufficiently bizarre that I'd just
as soon not type it in. In general, however, remember that
MAC address uniqueness is an especially hazardous assumption
in a multiprotocol routed environment. XNS and DECnet are
the likeliest to cause problems, but they are not the only
ones.
In non-routable protocol environments, such as NetBIOS/NetBEUI,
802.5 source route bridging, etc., administrators often
modify the MAC address to reflect the network topology.
Ongoing work in developing autoconfiguration mechanisms for
new-generation IP will probably depend on some unique identifier
for end hosts. That may be a MAC address, or something else.
IMHO, we need to get away from manually set MAC addresses.
Can anyone cite a need, other than compatibility with older
software, why we must have them?
Howard Berkowitz
PSC International
Follow-Ups:
References:
|
|
