jon@CS.UCL.AC.UK.UUCP (03/21/87)
We at UCL witness an interesting problem. Because it is regarded as AOK for gateways to talk to each other using unreliable IP, and gateways don't reassemble, only a minimal amount of info can get between gateways at any one go. The internet has got so large now, that when a reasonable percentage of sites are up, we disappear off the end of routing updates, and go unreachable. Why not get the gateways to reassemble packets destined for them (ie when they are acting as hosts/end points after all), I hear you cry? Nope, cos thats just a short term fix til there are more than (say) 1500 bytes worth of update. Why not use a transport protocol between IP and EGP/GGP/IGP/RIP etc? Yes, but which one. Well, there's TCP/RDP/ and who knows how many transaction protocols out there waiting to pounce. [This may also make your routing algorithms cleaner.] Well, use the same one as you use for talking management to your gateway from your hosts for now, like TCP. Surely it's not beyond the wit of gatewayfarers to put TCP into their boxes, since half of them build TCP terminal concentrators already. Our attitude to manageing MAC Bridges ~like~ the DEC LAN Bridge 100, is to put a separate management network (eg V24 or V35/ RS422 lines or whatever) in the ether/coax/fibre bundle, and use that to talk telnet/tcp blah into the Bridge. That's no solace to people with proprietry stuff in the Bridge, but I think it's the way things will go. Next years answer is: use ECMA ROS over REX, because it's gonna be a standard, and I am a biased European. Jon
hinden@ccv.bbn.COM (Robert Hinden) (03/23/87)
Jon, I don't think it OK for gateways to not do IP reassembly. They should be able to reassemble datagrams addressed directly to them. We are currently working to add this to the Butterfly and LSI-11 gateways. In principal I agree with you that we should think about using a transport protocol to carry our routing data. The problem I see is that most of our current protocols try very hard to deliver the old data before the new data. This is less than optimum for routing data, where it is important the deliver the new data and forget about delivering the old data. I hope we will never get to the point where we send our routing around by magtape like BITNET. Bob
jon@CS.UCL.AC.UK.UUCP (04/01/87)
Chris, ROS stands for Remote OperationS, and REX stands for Remote EXexcution. They are part of the ECMA (European Computer Manufacturers Association) input to draft standards work on distributed computing. REX is a Birrel & Nelson type 'minimum packet' transaction transport protocol, whilst ROS uses ASN (Abstract Sysntax Notation) as a presentation layer to to define call/reply/error messages and parameters and to provide machine independance of call and reply parameter representation. The ANSA (Advanced Network Systems Architecture) research group have put a lot of work into their design. The goal is to provide a system where code for client/server model interactions may be generated automatically (RPC type work), but to allow for other types of interaction too. jon