Re: Adjacency index

Bob Stewart <> Thu, 27 August 1992 17:59 UTC

Received: from by IETF.NRI.Reston.VA.US id aa06327; 27 Aug 92 13:59 EDT
Received: from NRI.NRI.Reston.Va.US by IETF.NRI.Reston.VA.US id ab06321; 27 Aug 92 13:59 EDT
Received: from by NRI.Reston.VA.US id aa16358; 27 Aug 92 14:01 EDT
Received: by; id AA01082; Thu, 27 Aug 92 11:00:35 -0700
Received: by; id AA12860; Thu, 27 Aug 92 10:42:04 -0700
Received: by; id AA12856; Thu, 27 Aug 92 10:42:03 -0700
Received: by; id AA00153; Thu, 27 Aug 92 10:41:59 -0700
Received: by id <>; Thu, 27 Aug 92 13:41:27 -0500
Date: Thu, 27 Aug 92 13:41:27 -0500
Message-Id: <>
From: Bob Stewart <>
In-Reply-To: John A. Shriver's message of Thu, 27 Aug 92 12:09:03 EDT <>
Subject: Re: Adjacency index

>The protocol is broken.  Just redefining the index the way Xyplex did
>does not solve the whole problem.  Then you have no idea which circuit
>an adjacency is on.  

Right.  I didn't think of that.  Actually, we may have done something truly
bizarre and indexed by a useful integer but returned the circuit index for the
allegedly corresponding MIB object.  Or maybe not.  That's academic.  In any
case the MIB is decidedly broken and it seems marking the table obsolete and
replacing it with the double indexed one is the right answer.

The encoding for the node address is indeed unfortunate.  It would be slick if
there was some way to keep it a single object and naturally divide it at the
right boundary, but I can't think of one.  Integer and octet string are both
relatively bad for a generic NMS, and separating area and node in all cases is
a pain.  Now if we wanted to consider that, we'd have to weigh the change
against existing implementations.  It's so much easier to design a proprietary
system, and that was no picnic.