rgt@lanl.gov (Richard Thomsen) (08/02/90)
Original-posting-by: rgt@lanl.gov (Richard Thomsen) Original-subject: Networking protocols at high speed (800 MBits/sec) Reposted-by: emv@math.lsa.umich.edu (Edward Vielmetti) This may not be the best group(s) to put this out on, so please email me some advice (no flames, OK?). There is an ANSI standards group (X3T9.3) that is working on a fibre channel that is to carry data at 800 MBits per second. It is commissioned to carry SCSI, IPI, and HIPPI on it at these speeds. It meets every month in both working groups and plenaries. It started as an extension of the HIPPI standards group, done in the same committee as HIPPI and IPI, and is designed to give HIPPI speeds to SCSI and IPI, and do all three on fibre. SCSI is in another X3T9 group, so is somewhat related. FDDI is in X3T9.5. HIPPI, for those that do not know, is a 32-bit parallel channel that will carry data at speeds of 800 MBits/second for up to 25 meters. The HIPPI physical standard is out for its second public review, and the next layers, including the 802.2 encapsulation layer, is currently in process. I and others desire to run networking protocols over this fibre channel. However, the committee is dominated by people with experience on channels to disks and tapes, and do not work the same way as networking protocols. For example, they do not like layering, and want physical circuit switched data transfer instead of datagrams. They are building a standard that will contain all information from the physical level (lasers and fibre) up to the application interface to SCSI, IPI, etc., although the document itself will be split into clauses that specify each "layer". It includes parts of the networking and transport layers. I am trying (with various levels of success) to get datagram service on this fibre channel. They have agreed to put datagram service on in general, although it is not exactly what I desired. I have also been repetedly pushed to figure out how to use their circuit based methods to do datagrams so they do not have to support datagrams directly. Here at Los Alamos, we are building a new network, in which we plan to run TCP/IP for now, with migration to OSI in the future. To do this, I am planning on putting both on 802.2, and putting 802.2 on the fibre channel. However, it is a very discouraging effort, and others have given up on fibre channel due to the domination by the channel people. If there is anyone on these newsgroups who is interested in running either TCP/IP and/or OSI on fibre channel, it may pay you to look into the fibre channel effort. With enough support, it may be possible to either sway the committee to a standard more conducive to this effort, or to come up with a different protocol on the basic physical layer that will be defined. In either case, there are those in the committee who are quite set in their ways, and hope to get this standard nailed down in the quite near future. It is almost too late to get things changed, but there is a little hope that things will improve. If anyone is interested in the fibre channel or HIPPI efforts, please send me email on the subject, since I do not read these newsgroups in general. It would be best if many networking people would come to the fibre channel meetings to see what is going on and to give their input. It seems to me that they are trying to solve problems that were solved in the networking efforts 25 years ago, but will not listen to me. Also, I have only a little experience in all this, so could use some help and advice. If a large number of manufacturers add fibre channel ports to their machines and workstations, it will be a channel that networks will wish to use, and it would be best to get your ideas in now. IBM seems to be very committed to the fibre channel project, and there are usually 9-12 IBM people from 5 IBM sites at the meetings. However, these are disk and tape people in general, not networking people. Other companies include Cray Research, CDC, Amdahl, DEC, LLNL, NCR, Sun, Unisys, etc. My last notes show 50 people from 29 companies attending. If desired, I can email a copy of the last minutes that I received on email, although I doubt they would make much sense without background information. Richard Thomsen rgt@lanl.gov