tkostan@TRWIND.TRW.COM (Tyson Kostan) (03/17/90)
I can give you a DoD view on the subject. I work for TRW and we are the supplier of TCP/IP componants to the Air Force under the ULANA (Unified Local Area Network Architecture) contract -- a requirements contract which will soon spread across all of the DoD. The Air Force has decided (or maybe better put, not decided) that they will do both. Under our contract, we will have to provide a manager which can speak both CMOT and SNMP. Our initial feeling on this subject was that this requirement came about due to 3COM's support of CMOT and not SNMP. I'm not too sure about this view at this time. 3COM claimed that it would have CMOT agents for all of its devices, but I'm not sure how far they have come with that. Now they are touting SNMP. DEC claims to be an ISO shop, and will be providing managers which support both CMOT and SNMP capabilities. I was reciently at COM-Net '90, and I didnt see any support for CMOT, where support for SNMP was everywhere, either in the form of managers, agents or proxy devices. In my opinion, ever since CMOT and SNMP went their own seperate ways (when the MIB split), SNMP has completely taken center stage, and bumped CMOT to the shadows. Only time will tell...
sean@dsl.pitt.edu (Sean McLinden) (03/17/90)
In article <9003170417.AA26725@psi.com> schoff@PSI.COM ("Martin Lee Schoffstall") writes: > >Now the question is will SNMP knock out CMIP itself in the area of NMS/agent >management within the OSI stack. I sincerely hope not. SNMP is probably adequate for the task of network management of networks and the full implementation of CMIP is, most likely, overkill for what are basic network management tasks, but when you start talking about the management of network objects that represent the state of the real world (I mean, the computers are there to support more than their network connections!) is when you see the real power of a protocol like CMIP. At the risk of sounding heretical, my personal opinion is that an implementation of CMOT would make TCP/IP sufficiently robust as to make it an attractive competitor to OSI (at least until we run out of addresses). To put this another way, the real appeal of OSI (and there are a lot of unappealing things about it), is not the market hype but what is the OSI concept of information as it exists in communities (an idea retrofitted to TCP/IP with such tools as XDR, RPC, Threads...). With a generalized interface to what is network information a la CM{IP,OT} you have a pretty powerful system which has the advantage of being widely available across multiple architectures. Until full implementations of the OSI stack become widely available CMOT would extend the functionality of TCP/IP as an OSI prototyping tool for, at least, the next few years, allowing for the applications to be developed in parallel with the network. Sean McLinden Decision Systems Laboratory University of Pittsburgh
mrose@CHEETAH.NYSER.NET (Marshall Rose) (03/20/90)
Well, I haven't seen the message you have replied to, but I guess I should speak up. If, for the moment, we forget the rather tainted history of CMIP in the Internet, the politics, failures, and so on, the issue still boils down to one of technical credibility. If you start comparing OSI application services and protocols (MHS, DS, FTAM, VT, CMIP), you see that the skill set is all over the board. Things like message handling and directory, whilst ambitious, work because they are basically solid protocols well-suited for a particular domain. In those two cases, I think the argument of "let's get it fully implemented and it will work out". In the case of FTAM and VT, their scope of focus is simply too large to be workable, e.g., the most charitable thing I've heard said about FTAM is that "it's both a dessert topping and a floor wax". The dessert topping part refers to record access (NFS-like) features and the floor wax part refers to bulk transfer (FTP-like) features. SNMP works well at network management because it is well-suited towards the task. When you need to cast lightning bolts, you conduct it through gold (SNMP) not paper (CMIP). This is akin to saying that SNMP is a network management protocol, and *perhaps* CMIP is a system management protocol, though I have no confidence in the latter portion since all the CMIP proponents I cross swords with keep talking about doing both, so they have fallen into the dessert topping/floor wax trap. As you might guess, anything that can be used as a floor wax probably doesn't taste too good, and anything that can be used as a dessert topping probably doesn't clean too well. But, that's CMIP for you. /mtr