daveb@gonzo.UUCP (Dave Brower) (11/26/88)
I would appreciate hearing (anecdotal) reports of the costs people have
observed running tcp/ip over international x.25 networks. I am
particularly interested in how the packet charges add up when there is
sporadic interactive use with telnet-like applications.
As background, there seem to be two ways to talk from here to there,
leased "line" and public x.25 network. With sufficient gateway
hardware, one can run tcp/ip over either, and the choice comes down to
cost. With a leased line you *know* what the cost will be: many
kilo-sheckels/month. With x.25 you are traffic sensitive, both by
volume and density. The worst case seems to be interactive login, where
it might take one (or two with echo) chargable packet per character.
Thanks,
-dB
--
If life was like the movies, the music would match the picture.
{sun,mtxinu,hoptoad}!rtech!gonzo!daveb daveb@gonzo.uucprhc%rcole@HPLB.HPL.HP.COM (Robert Cole) (11/28/88)
Dave, I can offer some reports, and I suggest you contact bruce@cs.ucl.ac.uk for more stories. I used to work at University College London where we used an international X.25 link to a gateway at BBN to transfer considerable traffic into the ARPANET. We used this line for several years (the original IP over X.25 spec was written for this application, it was only published when the CSNET folks needed it). > I would appreciate hearing (anecdotal) reports of the costs people have > observed running tcp/ip over international x.25 networks. I am > particularly interested in how the packet charges add up when there is > sporadic interactive use with telnet-like applications. I wrote a paper for a conference describing measurements taken on the link, and made some comments about the charging. The major point I made was that we were being ripped off! The international link is limited to 2 packet windows, however the round trip time of 1-2 seconds meant we were waiting a long time with a closed window. In the meantime our connect charges were running away. The PTTs were not sympathetic to this accusation and sent a top level delegation to the conference to defend their position. > As background, there seem to be two ways to talk from here to there, > leased "line" and public x.25 network. With sufficient gateway > hardware, one can run tcp/ip over either, and the choice comes down to > cost. With a leased line you *know* what the cost will be: many > kilo-sheckels/month. With x.25 you are traffic sensitive, both by > volume and density. The worst case seems to be interactive login, where > it might take one (or two with echo) chargable packet per character. We did run TELNET traffic over the X.25 link, but we insisted on line at a time working by imposing a terminal level relay that bundled stuff up to send. In fact most UK systems are line-at-a-time anyway. However, we couldn't stop the US hosts sending small packets, or blasting us with ACKs, for which we duly paid! It is surprising how much un-economic activity TCP implementations carry out when they are tuned for FREE networks. We generally felt that we would be better off with a leased line, but the PTT refused to supply one. THings have changed and the service is now moving (has moved) over to a 64K bps satellite link. I did keep a lot of charging information, but I don't know what happened to it, Bruce at UCL may be able to help you some more. Cheers, Robert PS The paper citation is: %A R. H. Cole %T User Experience and Evaluation of International X.25 Services %J Proc. Telecoms Today Conf. %I Online %C London %D March 1984 %P 107-118 %K X.25 performance If you have trouble getting a copy, send me a paper posting address and I will send you a copy.
craig@NNSC.NSF.NET (Craig Partridge) (11/28/88)
> I would appreciate hearing (anecdotal) reports of the costs people have > observed running tcp/ip over international x.25 networks. I am > particularly interested in how the packet charges add up when there is > sporadic interactive use with telnet-like applications. Dave: Two weeks ago I had the interesting experience of telneting in to read my mail in Boston from an NTT site in the Tokyo area. Regarding packet charges, two observations are likely to be of interest to you: (1) International X.75 gateways typically only allow X.25 windows of size 2. (2) Because of (1) the Nagle algorithm + type ahead really helps. The round-trip times were on the order of 20-30 seconds, so waiting for the first character to be echoed was silly. I just typed in my commands as fast as I could. When the ack for the first character came back, by typing ahead and allowing the Nagle algorithm to buffer up the data, I ensured that there was lots of data to go in the next packet. In other words, if you use the Nagle algorithm you should only very rarely see one character per packet. And if you don't use the Nagle algorithm you probably won't be able to use the connection at all -- 2 characters every 30 seconds just doesn't cut it, while once I got to typing ahead, I found the connection usable, if slow. Craig