[Simh] Comments WRT to IP vs ISO and OSI model

[Paul Koning]

> On Oct 6, 2015, at 9:08 PM, Clem Cole <clemc at ccc.com> wrote:
> 
> 
> On Tue, Oct 6, 2015 at 8:36 PM, Johnny Billquist <bqt at softjar.se> wrote:
> Right. I suspect Rich was actually thinking of the 7 layer OSI model, which DEC tried really hard to implement.
> ​insert >>eventually<< in there and I agree.
> 
> In the first few generations of DECnet, as Rich points out, DEC was no different than anyone else and had a closed networking system and it was not that clean.  By DECnet phase III many lessons had been learn and it really was a nice subsystem.  However, I've always felt that one of the failures of DECnet was the dogged adherence to ISO protocol later made them ignore the "IP" part of the the Internet protocol for too long - because in fact OSI did not really have it in its model.

You're mixing up layer models and specific protocols.  TCP/IP is (roughly) an instance of the OSI model, just as DECnet Phase III and IV are.  Specifically, DECnet routing layer, IP, and OSI IS/IS are all network layer (layer 3) protocols.

> By the time the ISO guys added intra-networking (network of networks - or what I referred to as "Dave Clark's observation"), Metcalfe's law (the value of a communications network is exponentally proportional to the number of things connected to it)​ it was too late.   And by then the US Gov had paid for enough IP/TCP implementations and there were so many on the Internet that even IBM and Microsoft could not catch up (although they too tried).

True.  What DEC missed is the speed of adoption of all the stuff that came with Unix, once Unix became a force in the commercial world (which, I believe, is due to the success of Sun).

> (And no, TCP/IP do *not* follow the 7 layer OSI model.)
> ​Amen, not by a >>long<< shot.   You could sort of map it, and we all tried to explain the IP stack in those terms, but you are so right.

I'm not sure anything ever precisely followed the 7 layer OSI model.  In any case, that's not particularly interesting.  The 7 layer model is nothing more than a guideline to protocol designers.  What actually matters in reality is the protocols.

TCP, like DECnet Phase III, has approximately 4 layers, which can be roughly mapped to the datalink, network, transport, and application layers of OSI.  Phase III NSP also includes the session layer; TCP never really had one (that's left to platform-specific machinery like inetd).  And no one really took the presentation layer seriously.  Also, in the real world the bottom layer generally contained both physical and data link layers.

DECnet Phase IV separated out the session layer in the specifications.  In the real world that didn't make any appreciable difference -- it slightly perturbed the NCP command syntax but apart from that it doesn't matter to anyone.

> That said, one of the problems with the OSI model was it did not do a good job with the "network of network" concept which was what made the Internet take off because it enabled Metcalfe's law to be in effect.
> 
> It seems to obvious today, but at the time, it was not so clear.  A lot of very smart people believed in closed networks.

Depends on how you mean closed networks.  OSI was meant to be open.  DECnet is also open, in the sense that Sun used the term -- DEC owned the specs, anyone is free to implement to the specs.

The trouble with the protocols (not the layer model) is internetworking.  IP had that as part of the concept from the start.  DECnet did not; it wasn't meant to go outside a single organization.  OSI *did* think about internetworking, but because of its history it thought about it the wrong way.  The origins of OSI are in the telco proprietary networking world -- the world of X.25.  This is where the standards describe only the customer interface, and what the telcos do internally is entirely proprietary, likely to vary from telco to telco.  Oh yes, there may be interworking specs, which specify what happens at the telco boundaries (if memory serves, that was X.75).  Also, the addressing definitely intended to support globally unique addresses -- just as, say, fully qualified telephone numbers are globally unique, as are IP addresses (ignoring NAT) but not DECnet Phase IV addresses.

Those origins explain why OSI protocols initially were connection oriented, again like X.25.  It wasn't until DEC and others like it got involved that OSI was forced into including datagram services, in the network layer specifically.  That also is when standardized routing protocols were created -- OSI IS/IS is largely the work of DEC and derived from an early proposal for DECnet Phase IV routing that was set aside ("too hard") in place of the "Phase III-E" hierarchical distance vector protocol we now know as Phase IV.  DEC also helped make the TP-4 transport layer work right for use on datagram network layers -- this was based on the experience with NSP and congestion control or congestion avoidance specifically.

But in all that, there wasn't enough recognition that IS/IS isn't good for internetworking; IP uses BGP there rather than OSPF for that reason.  IS/IS is like OSPF (in fact, it is the ancestor of OSPF).  There isn't any OSI protocol analogous to BGP.  I don't know why.  It's possible the remaining telco influence in the OSI effort prevented its creation.  Or it may be a lack of imagination.

	paul