ken@gvax.cs.cornell.edu (Ken Birman) (06/12/87)
Oops! I was indeed thinking of the 7-layer reference model, which is what I thought people were arguing for -- not the protocol suite, X.25, or whatever. Thanks for pointing this out. I find the multiple uses of ISO confusing. My feeling that multicasting and other higher-level functions (blast transfer protocols, etc) are in the long run going to be very important for building high performance distributed applications and especially robust applications, for example the fault-tolerant ones that my group is focused on. They may actually become more important in practical terms than point-to-point protocols of any kind. I appreciate that there is a lot to argue about regarding the relative inadequacies of TCP, XNS, X.25, or what have you. But, arguing these issues won't make it easier to build really sophisticated distributed software. What is needed is a "leap" in the level of abstraction to a kind of distributed computing environment in which this kind of application is just easier to build. Why do I consider this relevant? Well, it seems natural to me that before one decides which standard is "best", one should have a very good sense of what application programs will demand a decade from now. And, I don't feel that we have a good sense for this because all our experience is RPC oriented and stream oriented, and these may in the long run turn out to have been a very inadequate style of communication support. I would support this claim by observing that very little distributed software exists and what does exist is remarkably unsophisticated about distributed system "state". Moreover, very little of it is particularly robust or able to dynamically adapt to changing conditions. If, as I think, the real issues are at this higher level, than the kind of low level transport protocol one uses is probably not going to be TCP or X.25 (unless the world has standardized on one of these and there is no choice). It will be something else (VMTP?) optimized for the sorts of things the high level abstraction must support. So, I would propose that people discuss "high level requirements" for a while, in order to build up a strong argument for improving performance of those classes of applications that are poorly served by the existing protocol suites. From this, a better understanding of what the proponents of the different standards are really talking about might emerge, and then we would really be doing something productive... By the way, questions about the exact functioning of Dave Cheriton's task force are best addressed to Dave. I'm just a member. Dave is also on the end-to-end group, which I am not. So, I just can't answer these sorts of questions. Also, if you arer worrying about this, other members of the group seem far more knowledgable about the existing standards (ISO included) than I am. Ken