Comment Re:But OSI killed Decnet (Score 1) 375
Time to fill in some history from one who was in the middle of this. (I was part of the DEC network architecture group when the DECnet Phase V design was done.)
OSI is *not* just a model. The 7 layer reference model may be the best known part, but there is/was a whole collection of OSI protocol specs.
The 7 layers wasn't the problem. All sane protocols have layers. Maybe fewer (e.g., 4 or 5 in the TCP stack depending on how you count; 4 to 6 in the DECnet Phase 4 stack depending on how you count) or more. And the OSI insistence on abstracting the layer interfaces was a good thing, one since then widely adopted in other places. Also, in spite of Tanenbaum poking fun at it, terms like "PDU" are useful.
What hurt OSI was a bunch of things.
Much of its ancestry is X.25, a particularly obnoxious and ill-designed "network". Originally OSI was going to be connection oriented all the way down the layers. When DEC got involved, we beat some sense into them, so the connectionless network layer was added.
As a result, OSI ended up very large because lots of incompatible goals all had to be permitted. So the transport layer comes in five flavors, four of which are worthless. And the network layer comes in two flavors, one of which is a waste.
Since the work was done in an international standards committee, it moved very slowly. TCP/IP got the work done much faster. For example, few people know that OSPF is derived from the OSI IS/IS (router to router) protocol. OSPF came out first, but not because it was created first -- only because it went through a much faster process.
OSI the protocol stack is essentially dead. But pieces of it served as the basis for things that really matter. OSPF is a particularly good example. And OSI (preceded by DECnet) got some things right that TCP got wrong -- a packet interface rather than a byte stream interface, in particular. SCTP may one day fix that in the IP world, though I'm not holding my breath on that one. Until it does, everyone ends up recreating packets from the byte stream at the layer above TCP...
ATM doesn't "fit neatly on the bottom 3 layers of OSI" at all. It may have wanted to back when there was a dream of ATM all the way from the globe to the desktop, but that never was realistic, and in real life ATM is just an unusually complex datalink.
(Bruce Mann, inventor of LAT, once jokingly referred to the "49 layer model" because everyone seems to want to put the whole layer stack inside his own little piece of it. Then again, by inventing LAT he started that process...)
OSI is *not* just a model. The 7 layer reference model may be the best known part, but there is/was a whole collection of OSI protocol specs.
The 7 layers wasn't the problem. All sane protocols have layers. Maybe fewer (e.g., 4 or 5 in the TCP stack depending on how you count; 4 to 6 in the DECnet Phase 4 stack depending on how you count) or more. And the OSI insistence on abstracting the layer interfaces was a good thing, one since then widely adopted in other places. Also, in spite of Tanenbaum poking fun at it, terms like "PDU" are useful.
What hurt OSI was a bunch of things.
Much of its ancestry is X.25, a particularly obnoxious and ill-designed "network". Originally OSI was going to be connection oriented all the way down the layers. When DEC got involved, we beat some sense into them, so the connectionless network layer was added.
As a result, OSI ended up very large because lots of incompatible goals all had to be permitted. So the transport layer comes in five flavors, four of which are worthless. And the network layer comes in two flavors, one of which is a waste.
Since the work was done in an international standards committee, it moved very slowly. TCP/IP got the work done much faster. For example, few people know that OSPF is derived from the OSI IS/IS (router to router) protocol. OSPF came out first, but not because it was created first -- only because it went through a much faster process.
OSI the protocol stack is essentially dead. But pieces of it served as the basis for things that really matter. OSPF is a particularly good example. And OSI (preceded by DECnet) got some things right that TCP got wrong -- a packet interface rather than a byte stream interface, in particular. SCTP may one day fix that in the IP world, though I'm not holding my breath on that one. Until it does, everyone ends up recreating packets from the byte stream at the layer above TCP...
ATM doesn't "fit neatly on the bottom 3 layers of OSI" at all. It may have wanted to back when there was a dream of ATM all the way from the globe to the desktop, but that never was realistic, and in real life ATM is just an unusually complex datalink.
(Bruce Mann, inventor of LAT, once jokingly referred to the "49 layer model" because everyone seems to want to put the whole layer stack inside his own little piece of it. Then again, by inventing LAT he started that process...)
