A From:	SMTP%"RELAY-INFO-VAX@CRVAX.SRI.COM" 25-JUN-1993 11:45:03.97  To:	EVERHART CC:	/ Subj:	DECnet IV/V migration issues - a rebuttal   % Date: Thu, 24 Jun 1993 18:40:36 -0700 % From: Arun Sastry <asastry@cisco.com> 2 Message-Id: <199306250140.AA29838@regal.cisco.com>F To: knibbe@cisco.com, agt@cisco.com, gkunis@cisco.com, dkatz@cisco.comE Cc: afeather@cisco.com, dickey@cisco.com, williams_b@snd01.pcr.co.uk, 9         Info-VAX@kl.sri.COM, KEDDIE@sndsu1.sinet.slb.com,           larmer@nstg.enet.dec.com2 Subject: DECnet IV/V migration issues - a rebuttal    % 		Phase IV/V conversion and migration % 		===================================   H This is with regard to the talk I gave at DECUS in Atlanta two weeks ago on DECnet IV/V issues.  L It has come to my attention that some people have been having major concernsM about cisco's implementation of the IV/V conversion algorithm. There seems to M be a perception that cisco does not follow the algorithm that is specified in M DEC's PhaseV architecture and is somehow 'different'. This might be because I O did not emphasize the fact that our algorithms adhere to the specifications (as O outlined in DEC's Phase V architecture) and I focussed more on the differences   between the two philosophies.   L The differences between the two vendors' approaches  has to do with WHEN theM conversion is done. A DEC router running Phase V will always convert Phase IV H packets to Phase V i.e. conversion occurs as soon as possible. With thisJ scheme, a packet from a Phase IV host bound for another Phase IV host willJ undergo two conversions if it encounters a Phase V router - from IV to V, M then through the V cloud, and then back to IV when it exits the V cloud. This 3 is what I referred to as the "integrated" approach.   N Cisco routers on the other hand follow the "Ships-in-the-night" approach. WithK this scheme, a packet will be forwarded in its native format for as long as M possible, and will get converted only when there is no forwarding information J in the native format. In this case, a cisco router will convert the packetM and send it on if a route exists. The advantage of this scheme is that we can  avoid conversion in some cases.   M The other issue has to do with the way a cisco router with conversion enabled K deals with adjacencies. When a cisco router hears a IV hello, it creates an L Phase IV adjacency and also converts the address to Phase V and stores it inJ its Phase V data base. Similarly, a Phase V hello is converted to Phase IVK (if it is Phase-IV compatible) and stored in the Phase IV data base. On the J other hand, a DEC router running Phase V will convert a Phase IV hello andH store it in its Phase V data base. There is no duplication of end-systemN information in a DEC router. I also spoke about the fact that DEC architectureM specifies that an area must be either Phase IV or V, while cisco's allows you L to have both IV and V in some (or all) routers. The advantage of this schemeJ is that there is no need to convert the entire backbone to Phase V, and so) allows a more gradual migration strategy.   J I then spoke about a new feature that we were coming out with in the next L software release. This had to do with being able to propagate Phase IV area O information through a Phase V cloud. Currently, reachability to remote Phase IV O *AREAS* is not "redistributed" through OSI, so that IV nodes on one side of the O cloud will not hear about other Phase IV areas hanging off of the router at the L other end. The scheme I outlined was a static one, since Phase IV areas thatO needed to be advertised had to be statically entered. This was a first step in  M overcoming this limitation. In connection with this, I also mentioned that a  L more elegant scheme would be to have a dynamic IV/V route redistribution. ItN is important to note here that what this capability gives you is to be able toO advertise ADDITIONAL Phase IV AREAS across the cloud; we have always been able  : to propagate level-1 reachability through a Phase V cloud.  H Let me reiterate again that the translation algorithms followed by both O vendors' implementations are the same, and there should be no interoperability   problems arising from this.   K I apologize if I did not present the facts more clearly and if I raised any L red flags. Furthermore I apologize if I have misunderstood the concerns you N were trying to raise, and if so, I will be happy to clarify them and put them  to rest.  J Please send all mail to cs-decnet@cisco.com so that it will be seen by all the right people.    Regards,  
 - Arun Sastry      Software Engineer    cisco Systems    1525 O'Brien Drive   Menlo Park, CA 94025                 