db@witzend.East.Sun.COM (David Brownell) (07/14/90)
In article <2510@sequent.cs.qmw.ac.uk> liam@cs.qmw.ac.uk (William Roberts) writes: > I still think that in terms of having a better NFS in the next > year or so, you would be better implementing a named extensions > mechanism similar to that used by X. This has the advantage > that small extensions can be tried and ractical experiences > folded into the grand NeFS redesign, rather than having to get > all of the NeFS design right first time. In effect, this mechanism exists today. The RPC portmapper is the mechanism for naming the extensions, and querying whether they exist. It's immaterial that it names other protocols, as well as the NFS-friendly ones. The NFS protocol in essence consists of four subprotocols: Mount, NFS, Lock Manager, and Status Monitor. The last two don't get much press of late, but if you try writing a DOS client (for example) you'll require them else you can't host DOS file systems. (OSF has in essence said that the PC-NFS daemon runs a fifth subprotocol; it's in OSF/1.) One serious problem with this kind of mechanism is that if this is the only way to extend your system, you quickly run into some interesting compatibility problems on Real Networks (tm). Those five subprotocols come in three groups (core, LockMgr/StatMon, PC-NFSd) and just how is a site supposed to cope with Vendor X only supplying the core? Much better to solve this particular software distribution problem by eliminating it; it can only get worse if more extensions get defined. By the way, the X community is hitting situations where this mechanism doesn't really hit the spot -- for the same configuration management reasons. If an application relies on the SHAPE extension, and the server doesn't support it, the application can't run. The customer who bought the application loses. I dislike postscript programming as much as the next guy; I'd rather NeFS used a language like Scheme, which is easily compiled and allows lots more error checking. But as someone who's had to deal with upgrading distributed systems, I really like the idea of low level system support for it that NeFS proposes. #include <std/disclaimer.h> David Brownell db@east.sun.com. "What's the network equivalent of 'the rough section of town'?"