Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!wuarchive!uunet!rufus!drake.almaden.ibm.com!drake
From: drake@drake.almaden.ibm.com
Newsgroups: comp.unix.aix
Subject: Re: strange tcpip problem:  SOLVED!!
Message-ID: <843@rufus.UUCP>
Date: 13 Jun 91 06:48:15 GMT
References: <1991Jun12.230426.22643@milton.u.washington.edu>
Sender: news@rufus.UUCP
Organization: IBM Almaden Research Center
Lines: 30

In article <1991Jun12.230426.22643@milton.u.washington.edu> eliot@engr.washington.edu (Eliot Lim) writes:
>I don't know if this is a bug or a feature, but if you move your ethernet
>card, the OS will assign it a new hardware address and leave the old one
>and treat it as a device that has gone offline.  

Feature.  How else could it work?  Consider a scenario ...

Machine has two Ethernet adapters, one in slot 3 attached to network A,
and one in slot 4 attached to network B.  Lightning strikes and the
adapter in slot 3 goes "poof".

Now, when the machine comes up, I really want it to remember that the
slot 4 Ethernet card is on network B, not go playing guessing games.
Any method that didn't make the adapter definitions slot dependent
would cause icky problems in such scenarios...you'd wind up with an
adapter on network B, perhaps with a definition for network A ... total
chaos.  Similarly (and identically from the software's viewpoint),
pulling out the slot 3 card shouldn't screw up the connection to
network B.

SO, the moral is that while all slots are created equally, the slot
address of a card is just as important and unchangeable as the SCSI
address of a drive.

How do other systems handle this?


Sam Drake / IBM Almaden Research Center 
Internet:  drake@ibm.com            BITNET:  DRAKE at ALMADEN
Usenet:    ...!uunet!ibmarc!drake   Phone:   (408) 927-1861