Re: Various DECnet MIB questions & Adjacency index

EMA - we manage almost anything! <sylor@blumon.enet.dec.com> Fri, 28 August 1992 19:40 UTC

Received: from ietf.nri.reston.va.us by IETF.NRI.Reston.VA.US id aa05854; 28 Aug 92 15:40 EDT
Received: from NRI.NRI.Reston.Va.US by IETF.NRI.Reston.VA.US id aa05850; 28 Aug 92 15:40 EDT
Received: from inet-gw-2.pa.dec.com by NRI.Reston.VA.US id aa14829; 28 Aug 92 15:42 EDT
Received: by inet-gw-2.pa.dec.com; id AA17389; Fri, 28 Aug 92 12:37:51 -0700
Received: by nsl.pa.dec.com; id AA12177; Fri, 28 Aug 92 12:33:52 -0700
Received: by nsl.pa.dec.com; id AA12043; Fri, 28 Aug 92 12:10:42 -0700
Message-Id: <9208281910.AA12043@nsl.pa.dec.com>
Received: from blumon.enet; by nsl.enet; Fri, 28 Aug 92 12:33:42 PDT
Date: Fri, 28 Aug 92 12:33:42 PDT
From: EMA - we manage almost anything! <sylor@blumon.enet.dec.com>
To: phiv-mib@pa.dec.com
Cc: sylor@blumon.enet.dec.com
Apparently-To: phiv-mib@Pa.dec.com
Subject: Re: Various DECnet MIB questions & Adjacency index

> I'd at least ask that we generate and document a rule as to which route is
> returned, if there are multiple possibilities.  How about lowest circuit
> index first, followed by lowest next-hop address for multiple choices on
> the same circuit.
> 
> Art

    Art brings up a good point. In a multiple path network, you'ld like some
    consistency in which path is shown. The "first circuit", and then the
    "lowest number" adjacency are good choices.

    But that reminds me of a bit of folklore that you might want to know.

    It is important that the "first circuit" be deterministic. As long as the
    router's configuration doesn't change, the indexes of the circuits should be
    the same. And it would be good if the order of the circuits by index was
    the same as the alphabetic order of the circuits by name.

    There is a requirement that table indices be stable and deterministic in
    one of the SNMP specs, I forget which. But the determinism I'm looking for
    is not just a requirement for network management, it is also a requirement
    for correct operation of routing. 

    We ran into a problem years ago when an implementation of DECnet indexed
    the circuits based on the order they came up. Well, given phases of the
    moon and other variables, the circuits ended up having a different order
    each time the router booted. That changed the whole routing pattern in the
    network (this was before path splitting) since if a router had 2 paths of
    equal cost, on Monday all packets would go left, and Tuesday all packets
    would go right, even though the network global configuration was exactly
    the same. Gave us no end of trouble in trying to come up with a workable
    network design or in diagnosing problems with "flaky routers". 

    Because of that experience, we became pretty fanatical about making routing
    deterministic. The most visible example of this is the "tie breaking"
    algorithm in chosing a dedicated router based on highest ID.

    Well, there's some free advice.

    					Mark