enger%gburg.DECnet@BLUTO.SCC.COM.UUCP (11/30/87)
Gentlepersons: CONTEL Federal Systems became the first foster parent in the Adopt-A- Mailbridge campaign. This campaign is an outgrowth of the suggestion to upgrade the EGP core gateways which was made at the recent IETF meeting. The EGP upgrade appears to have been very successful, reducing delays and packet loss by up to two orders of magnitude while the network load increased. The Adopt-A-Mailbridge program, as I've coined it, encourages members of the Internet community to improve service for all by making a temporary loan of equipment to upgrade a mailbridge gateway. Contel loaned an 11/73 processor to DCEC-MILNET-GW.ARPA. This note lists the results of some informal ping testing of the upgraded DCEC-MILNET-GW.ARPA (10.7.0.20), in comparison to concurrent tests of some of the other mailbridge gateways, and against a few tests of 10.7.0.20 taken before the upgrade. I do not have access to a net-26 host with which I could have performed throughput testing, so ping echo results are all I can provide. When comparing the performance of the upgraded DCEC to the other mailbridges keep in mind that more PSN hops must be traversed on net-10 to get to the other mailbridges. The host that I performed the testing from is connected to PSN DCEC20, as is DCEC-Milnet-GW.arpa. The "testing" consisted of pinging the net-10 interface address of the various mailbridge gateways and recording the average echo delay. Most measurements are based on receiving approximately 100 replies. One measurement of the original DCEC mailbridge received only 47 replies. For the purpose of removing the net-10 delay involved in opening a connection to the distant end, each test was preceded by a separate ping session of sufficient duration to see a few responses come back. Comparing the longest average delay measured from DCEC mailbridge (347ms) to the shortest average from any of the others (682ms), the improvement is a factor of 2. Comparing the average of all the upgraded DCEC measurements (279), to the average of the average measurements from all the others (1034), the improvement is a factor of 3.7. It would be useful if the effects of the extra net-10 PSN hops could be removed. Perhaps a conservative way to take a swag at this is to subtract off the afternoon net-10 average end to end delay (422ms) from the average delay figure for the remote mailbridges. This yields a corrected (??) delay figure for the remote mailbridges of 612ms, and an improvement factor of 2.2. My only data for the old DCEC was taken on the weekday test of the new PSN end to end software. Four measurements were taken. The longest of these received 129 packets, and lost 31%. It had an average delay of 14479ms. I am not considering this measurement. The average of the remaining three average delay measurements is 803ms. Based on this, the upgraded DCEC running on the old end to end is 2.88 times quicker. While the improvement is not as dramatic as that seen on the Arpanet EGP core gateways, it is still significant. I think we should try to locate Foster Parents for the rest of the mailbridges. Steve Atlas also reminds me that the EGP core gateways on the Milnet are still "un-sponsored". Finally, I wonder if the EGP core gateways would benefit from an even faster processor? Does anyone know if we can drop in an 11/83 cpu as easily as we dropped in the 11/73s? Has anyone got one to loan out? I hope all of you had a nice Thanksgiving Day, Bob Enger Contel Federal Systems enger@bluto.scc.com
Mills@UDEL.EDU (11/30/87)
Bob, I love your foster-gateway program. Turns out 11/83s would require special memory boards and backplanes, so would probably not make good orphans. Dave