Re: Single subnet

Dave Katz <katz@merit.edu> Mon, 21 May 1990 13:53 UTC

Received: from merit.edu by NRI.NRI.Reston.VA.US id aa23914; 21 May 90 9:53 EDT
Received: Mon, 21 May 90 08:52:40 EST by merit.edu (5.59/1.6)
Date: Mon, 21 May 90 08:52:40 EST
From: Dave Katz <katz@merit.edu>
Message-Id: <9005211352.AA08847@merit.edu>
To: deering@pescadero.stanford.edu
Subject: Re: Single subnet
Cc: fddi@merit.edu
Status: O

>Sorry, I wasn't very clear.  If a load-splitting forwarding algorithm
>(which is suggested, but not required, by the IS-IS spec) is causing half
>of the packets to disappear into a black hole, a transport protocol such
>as TP-4 *might* be able to compensate (depending on the particular loss
>pattern), albeit with a severe degradation of performance.  Obviously,
>it would be better for the routing/forwarding algorithm to avoid such
>a failure mode, than to rely on the transport layer to survive it.
 
Ah, now I see what you mean.  Actually, I misspoke--the maintenance of
equal-cost paths by IS-IS takes place even for the last hop (IS to ES),
since the SPF algorithm runs based on adjacencies (of which there would
be two for dual-MAC systems on an unwrapped ring).  I was confused by
OSPF and dual IS-IS, which route to subnets rather than hosts (and
last hop routing is done by ARP, etc., rather than the IGP).  This whole
mess gets a little clearer every day...(and a little scarier).