bzs@BU-CS.BU.EDU (Barry Shein) (08/17/87)
The technical discussions are interesting in regards to things like statelessness vs statefulness etc. What the casual or more product oriented person needs to know is: 1. NFS has been delivered in production systems for years now, it works. 2. NFS is now widely available from vendors for heterogeneous networks running on many different machines compatibly. 3. NFS uses the Internet protocol suite for its lower layers, by and large your knowledge and current investment in internet hardware and software will help you set up and maintain an NFS network. None of the above is really true about RFS (except that it's probably available on some very small set of machines from one vendor.) By and large it's vaporware in its promises to be a widely available heterogeneous product and its implementation on a non-internet standard makes it harder to see what its potential utility is (you want to become a 3Bnet expert on Starlan twisted pair? why?) Let's put it this way, at Boston University we run a lot of NFS sites in different depts. We've been using it to share single copies of things like font directories, software and source distributions, data bases and network news all over campus and to combine clusters of machines into one single environment where users maintain one home directory wherever they log in. RFS can do all this in theory but you'll have to settle on a very limited (1?) set of machine types, it's the old tail wagging the dog. NFS works, we've had it in production for over a year and the NFS community here just grows and grows. Like I said, technical discussions interest me also, but as far as I can see there is a standard in heterogeneous network file systems: NFS. And about 100 vendors agree with me. RFS may have its pluses but in terms of being a viable product it strikes me as being too little too late. -Barry Shein, Boston University