barns@GATEWAY.MITRE.ORG (Bill Barns) (09/26/88)
I would like to hear what algorithms people are using to decide when to multiplex Transport connections on Network connections. The case of most immediate interest to me is TCP/IP over DDN Standard X.25 connections. As far as I know, there is no standard or published guideline which would stop me from multiplexing everything (to the same Network-level destination, of course) on a single X.25 connection; at the other extreme, I might open a new X.25 connection for each datagram. The same issue arises with ISO TP over X.25. Does it follow that the same algorithm is "best" for both cases? The case of ISO CLNP over X.25 seems to combine the two cases above. This is what I understand GOSIP to specify for the future, so this case is also interesting. I've heard that some problems in the ARPANET transition to PSN 7.0 had to do with this topic. So I imagine there are some interesting issues somewhere underlying the decisions an implementor might make, and perhaps some interoperation problems with some choices. I'd like to get a handle on this whole area. You're all invited to describe the behavior of existing implementations or your conception of the Right Thing To Do. Thanks in advance for all info, insights, comments, and suggestions... Bill Barns / MITRE-Washington / barns@gateway.mitre.org
malis@OAKLAND.BBN.COM (Andy Malis) (09/27/88)
Bill, For IP over DDN Standard X.25, the most efficient (in terms of PSN and network resources) algorithm is to multiplex all datagrams to the same network-level destination over the same X.25 VC, requesting the largest packet size (up to 1024 octets) supported by your host. This is the ONLY algorithm allowed by the old End-to-End, which is still in use on the MILNET. This also most closely emulates the network transport of AHIP (1822) messages, except that the window size is 7 rather than 8. If you are on a network running the new End-to-End (the ARPANET), your host is allowed to open more than one Standard connection to other hosts. You may wish to have your host measure the average utilization of its host access line, and if it has a combination of low access line utilization and a long queue of datagrams to a particular destination host, then open a second VC to that host. Note that opening multiple VCs to a destination host only wastes resources in the PSN, network, and presumably the host, if both of the above conditions do not apply. The maximum number of VCs that any host may open at any one time is configured by the network administration, so opening multiple VCs to one destination may make it harder for your host to open VCs to other destinations without having to clear previously open VCs. Also note that the PSN clears Standard X.25 VCs that have been idle for a period of time. This period is also configurable by the network administration. Regards, Andy Malis BBNCC PSN Development