Mills@UDEL.EDU (12/16/87)
Folks, As I have been heard to mutter from time to time (pun), watching local clocks reveals all kinds of interesting information about network congestion and routing flaps. At the end of this message is a summary of clock flaps recorded over the last couple of weeks at fuzzball host udel2, which is synchronized via the Network Time Protocol (NTP) to WWVB-clock fuzzball umd1 across 1822 gateway dcn-gw, ARPANET and one or the other X.25 gateways at U Maryland. At issue are the Maryland gateways, which carry a lot of traffic between the NSFNET and ARPANET communities and are vulnerable to X.25 virtual-cirkosis, which occurs when the virtual-circuit demand exceeds the maximum of 64 supported by their ACC 5250 interface and driver. Under conditions of severe congestion packets may be dropped or delayed for relatively long periods, which mean relative clock offsets measured by NTP will become unstable, exceeding even the cope of the sophisticated median-filtering and smoothing algorithms used by the fuzzballs. When this happens a timewarp (step-change in local time) occurs, which is then logged for later analysis. It should be understood that timewarps over noncongested paths, including most NSFNET, ARPANET and local paths, are extremely rare and, when they do occur, very serious congestion, delay spikes and/or black holes are occuring. Over the last two weeks timewarps between udel2 and umd1 have been occuring with increasing frequency. Casual ping reveals occasional severe congestion and packet loss across the Maryland gateways, but relatively smooth sail on other paths within the NSFNET, ARPANET and local sloughs. The pattern of congestion suggests that black holes pop up between the gateways and specific destinations and then disappear after a few minutes. This is exactly the pattern expected when virtual-cirkosis strikes. Black holes would be expected to affect low-traffic hosts more than high-traffic ones, which tend to recapture virtual circuits more quickly. This problem threatens to become the single most damaging factor affecting quality service between the NSFNET and ARPANET communities. Warnings about the problem have been issued on a regular basis since early October, when the problem first showed up at another gateway. There is no doubt that its incidence can be reduced by spreading the NSFNET/ARPANET load over more gateways and, in fact, additional gateways are already in place, but not configured for this function. Clearly the ACC 5250 and/or driver must be overhauled to provide either more circuits or more creative management of the existing number. Solutions for the driver problem, which involve cache-management and replacement strategies familiar to any student of computer science, have not been forthcoming. There is very real need to capture resources, voluntary or coerced, to assist in the solution. The data below will give some idea of the incidence of the problem, time of occurance and severity. It may be possible for other users to correlate these data with their local logs and trouble reports to see if the pattern matches. Remember, these data reveal only the most severe disturbances. Many, less severe disturbances are also affecting connections. 20:37:22 012 ?TRAP-I-Clock reset 4 1130 30-Nov-87 20:46:26 012 ?TRAP-I-Clock reset 4 846 30-Nov-87 21:52:41 012 ?TRAP-I-Clock reset 4 1000 30-Nov-87 22:01:48 012 ?TRAP-I-Clock reset 4 653 30-Nov-87 4 flaps 4:32:12 012 ?TRAP-I-Clock reset 4 920 02-Dec-87 4:55:17 012 ?TRAP-I-Clock reset 4 659 02-Dec-87 5:22:33 012 ?TRAP-I-Clock reset 4 641 02-Dec-87 22:24:29 012 ?TRAP-I-Clock reset 4 992 02-Dec-87 22:33:14 012 ?TRAP-I-Clock reset 4 634 02-Dec-87 23:28:00 012 ?TRAP-I-Clock reset 4 737 02-Dec-87 6 flaps 5:26:20 012 ?TRAP-I-Clock reset 4 637 03-Dec-87 5:28:52 012 ?TRAP-I-Clock reset 4 1200 03-Dec-87 6:33:57 012 ?TRAP-I-Clock reset 4 627 03-Dec-87 6:36:36 012 ?TRAP-I-Clock reset 4 628 03-Dec-87 7:35:26 012 ?TRAP-I-Clock reset 4 1188 03-Dec-87 7:46:40 012 ?TRAP-I-Clock reset 4 647 03-Dec-87 12:46:36 012 ?TRAP-I-Clock reset 4 1088 03-Dec-87 12:55:39 012 ?TRAP-I-Clock reset 4 639 03-Dec-87 16:10:01 012 ?TRAP-I-Clock reset 4 937 03-Dec-87 22:11:06 012 ?TRAP-I-Clock reset 4 753 03-Dec-87 22:29:25 012 ?TRAP-I-Clock reset 4 691 03-Dec-87 22:37:21 012 ?TRAP-I-Clock reset 4 659 03-Dec-87 23:30:55 012 ?TRAP-I-Clock reset 4 717 03-Dec-87 13 flaps 2:35:37 012 ?TRAP-I-Clock reset 4 1212 04-Dec-87 2:35:56 012 ?TRAP-I-Clock reset 4 674 04-Dec-87 3:02:18 012 ?TRAP-I-Clock reset 4 785 04-Dec-87 3:33:11 012 ?TRAP-I-Clock reset 4 873 04-Dec-87 3:35:49 012 ?TRAP-I-Clock reset 4 6527 04-Dec-87 3:35:59 012 ?TRAP-I-Clock reset 4 872 04-Dec-87 3:36:31 012 ?TRAP-I-Clock reset 4 11793 04-Dec-87 3:36:41 012 ?TRAP-I-Clock reset 4 829 04-Dec-87 3:39:03 012 ?TRAP-I-Clock reset 4 810 04-Dec-87 4:48:35 012 ?TRAP-I-Clock reset 4 741 04-Dec-87 4:52:18 014 ?TRAP-I-Clock reset 3 1000 04-Dec-87 7:36:38 012 ?TRAP-I-Clock reset 4 675 04-Dec-87 18:17:08 012 ?TRAP-I-Clock reset 4 680 04-Dec-87 19:22:46 012 ?TRAP-I-Clock reset 4 663 04-Dec-87 14 flaps 1:40:42 012 ?TRAP-I-Clock reset 4 649 05-Dec-87 5:12:26 012 ?TRAP-I-Clock reset 4 660 05-Dec-87 5:19:16 012 ?TRAP-I-Clock reset 4 633 05-Dec-87 17:04:36 012 ?TRAP-I-Clock reset 4 660 05-Dec-87 (Saturday) 4 flaps 21:41:11 012 ?TRAP-I-Clock reset 4 3409 06-Dec-87 21:50:22 012 ?TRAP-I-Clock reset 4 832 06-Dec-87 23:55:00 012 ?TRAP-I-Clock reset 4 2142 06-Dec-87 (Sunday) 3 flaps 0:04:26 012 ?TRAP-I-Clock reset 4 652 07-Dec-87 1:24:14 012 ?TRAP-I-Clock reset 4 3407 07-Dec-87 1:33:22 012 ?TRAP-I-Clock reset 4 646 07-Dec-87 2:38:18 012 ?TRAP-I-Clock reset 4 2308 07-Dec-87 2:47:27 012 ?TRAP-I-Clock reset 4 665 07-Dec-87 3:48:37 012 ?TRAP-I-Clock reset 4 2844 07-Dec-87 3:58:05 012 ?TRAP-I-Clock reset 4 638 07-Dec-87 8:20:53 012 ?TRAP-I-Clock reset 4 3185 07-Dec-87 8:30:38 012 ?TRAP-I-Clock reset 4 622 07-Dec-87 9:20:32 012 ?TRAP-I-Clock reset 4 3310 07-Dec-87 9:29:59 012 ?TRAP-I-Clock reset 4 657 07-Dec-87 17:28:46 012 ?TRAP-I-Clock reset 4 405 07-Dec-87 17:29:18 014 ?TRAP-I-Clock reset 3 1000 07-Dec-87 18:47:11 012 ?TRAP-I-Clock reset 4 1624 07-Dec-87 18:55:58 012 ?TRAP-I-Clock reset 4 656 07-Dec-87 19:30:38 012 ?TRAP-I-Clock reset 4 926 07-Dec-87 20:12:21 012 ?TRAP-I-Clock reset 4 654 07-Dec-87 20:24:29 012 ?TRAP-I-Clock reset 4 1081 07-Dec-87 20:33:43 012 ?TRAP-I-Clock reset 4 636 07-Dec-87 19 flaps 16:08:28 012 ?TRAP-I-Clock reset 4 1547 08-Dec-87 16:16:37 012 ?TRAP-I-Clock reset 4 666 08-Dec-87 2 flaps 3:43:00 012 ?TRAP-I-Clock reset 4 1159 09-Dec-87 3:52:06 012 ?TRAP-I-Clock reset 4 665 09-Dec-87 7:22:50 014 ?TRAP-I-Clock reset 3 1000 09-Dec-87 7:24:07 012 ?TRAP-I-Clock reset 4 1263 09-Dec-87 7:33:04 012 ?TRAP-I-Clock reset 4 658 09-Dec-87 16:16:41 014 ?TRAP-I-Clock reset 3 1000 09-Dec-87 16:37:08 012 ?TRAP-I-Clock reset 4 1234 09-Dec-87 16:46:53 012 ?TRAP-I-Clock reset 4 660 09-Dec-87 17:27:01 012 ?TRAP-I-Clock reset 4 1261 09-Dec-87 17:35:10 012 ?TRAP-I-Clock reset 4 661 09-Dec-87 19:33:53 012 ?TRAP-I-Clock reset 4 1419 09-Dec-87 19:42:07 012 ?TRAP-I-Clock reset 4 967 09-Dec-87 19:52:34 012 ?TRAP-I-Clock reset 4 667 09-Dec-87 20:06:59 012 ?TRAP-I-Clock reset 4 1389 09-Dec-87 20:15:27 012 ?TRAP-I-Clock reset 4 634 09-Dec-87 15 flaps 1:19:09 012 ?TRAP-I-Clock reset 4 1200 10-Dec-87 1:28:19 012 ?TRAP-I-Clock reset 4 662 10-Dec-87 17:05:09 012 ?TRAP-I-Clock reset 4 669 10-Dec-87 21:37:20 012 ?TRAP-I-Clock reset 4 1223 10-Dec-87 21:48:40 012 ?TRAP-I-Clock reset 4 677 10-Dec-87 22:05:13 012 ?TRAP-I-Clock reset 4 3457 10-Dec-87 22:13:55 012 ?TRAP-I-Clock reset 4 673 10-Dec-87 7 flaps 1:59:20 012 ?TRAP-I-Clock reset 4 1668 11-Dec-87 2:07:31 012 ?TRAP-I-Clock reset 4 653 11-Dec-87 2:25:09 012 ?TRAP-I-Clock reset 4 1299 11-Dec-87 2:33:59 012 ?TRAP-I-Clock reset 4 650 11-Dec-87 4:43:39 012 ?TRAP-I-Clock reset 4 2021 11-Dec-87 4:52:44 012 ?TRAP-I-Clock reset 4 917 11-Dec-87 6:19:59 012 ?TRAP-I-Clock reset 4 2515 11-Dec-87 6:28:47 012 ?TRAP-I-Clock reset 4 1038 11-Dec-87 8 flaps 3:12:21 012 ?TRAP-I-Clock reset 4 892 12-Dec-87 3:23:19 012 ?TRAP-I-Clock reset 4 2437 12-Dec-87 3:29:57 012 ?TRAP-I-Clock reset 4 1269 12-Dec-87 3:39:08 012 ?TRAP-I-Clock reset 4 724 12-Dec-87 5:43:07 012 ?TRAP-I-Clock reset 4 629 12-Dec-87 5:49:06 012 ?TRAP-I-Clock reset 4 1134 12-Dec-87 6:11:27 012 ?TRAP-I-Clock reset 4 3019 12-Dec-87 6:11:45 012 ?TRAP-I-Clock reset 4 3020 12-Dec-87 6:18:56 012 ?TRAP-I-Clock reset 4 1251 12-Dec-87 6:27:43 012 ?TRAP-I-Clock reset 4 662 12-Dec-87 9:10:28 012 ?TRAP-I-Clock reset 4 1205 12-Dec-87 9:19:34 012 ?TRAP-I-Clock reset 4 696 12-Dec-87 14:01:31 012 ?TRAP-I-Clock reset 4 1101 12-Dec-87 14:10:35 012 ?TRAP-I-Clock reset 4 629 12-Dec-87 22:00:22 012 ?TRAP-I-Clock reset 4 1047 12-Dec-87 22:10:44 012 ?TRAP-I-Clock reset 4 722 12-Dec-87 23:07:13 012 ?TRAP-I-Clock reset 4 926 12-Dec-87 17 flaps 2:46:49 012 ?TRAP-I-Clock reset 4 1435 13-Dec-87 2:54:56 012 ?TRAP-I-Clock reset 4 647 13-Dec-87 5:20:21 012 ?TRAP-I-Clock reset 4 959 13-Dec-87 5:28:10 012 ?TRAP-I-Clock reset 4 871 13-Dec-87 5:37:37 012 ?TRAP-I-Clock reset 4 1201 13-Dec-87 5:49:14 012 ?TRAP-I-Clock reset 4 644 13-Dec-87 22:59:57 012 ?TRAP-I-Clock reset 4 2039 13-Dec-87 23:13:06 012 ?TRAP-I-Clock reset 4 819 13-Dec-87 23:24:59 012 ?TRAP-I-Clock reset 4 1316 13-Dec-87 23:47:33 012 ?TRAP-I-Clock reset 4 674 13-Dec-87 10 flaps 0:23:00 012 ?TRAP-I-Clock reset 4 1254 14-Dec-87 0:31:48 012 ?TRAP-I-Clock reset 4 1577 14-Dec-87 0:40:18 012 ?TRAP-I-Clock reset 4 669 14-Dec-87 1:34:40 012 ?TRAP-I-Clock reset 4 1375 14-Dec-87 1:43:28 012 ?TRAP-I-Clock reset 4 652 14-Dec-87 2:37:20 012 ?TRAP-I-Clock reset 4 3142 14-Dec-87 2:45:50 012 ?TRAP-I-Clock reset 4 656 14-Dec-87 2:46:09 012 ?TRAP-I-Clock reset 4 657 14-Dec-87 5:07:25 014 ?TRAP-I-Clock reset 3 1000 14-Dec-87 5:21:47 012 ?TRAP-I-Clock reset 4 650 14-Dec-87 15:53:09 012 ?TRAP-I-Clock reset 4 1231 14-Dec-87 16:04:25 012 ?TRAP-I-Clock reset 4 695 14-Dec-87 16:37:40 012 ?TRAP-I-Clock reset 4 1304 14-Dec-87 16:48:06 012 ?TRAP-I-Clock reset 4 889 14-Dec-87 16:59:06 012 ?TRAP-I-Clock reset 4 1855 14-Dec-87 17:22:23 014 ?TRAP-I-Clock reset 3 1000 14-Dec-87 19:49:59 012 ?TRAP-I-Clock reset 4 1486 14-Dec-87 19:58:27 012 ?TRAP-I-Clock reset 4 1000 14-Dec-87 18 flaps 1:22:00 012 ?TRAP-I-Clock reset 4 1508 15-Dec-87 1:29:50 012 ?TRAP-I-Clock reset 4 660 15-Dec-87 2:44:13 012 ?TRAP-I-Clock reset 4 1429 15-Dec-87 2:52:22 012 ?TRAP-I-Clock reset 4 1007 15-Dec-87 15:21:31 012 ?TRAP-I-Clock reset 4 1454 15-Dec-87 15:29:41 012 ?TRAP-I-Clock reset 4 665 15-Dec-87 16:51:20 012 ?TRAP-I-Clock reset 4 1724 15-Dec-87 17:00:28 012 ?TRAP-I-Clock reset 4 1060 15-Dec-87 17:58:35 012 ?TRAP-I-Clock reset 4 1431 15-Dec-87 18:08:36 012 ?TRAP-I-Clock reset 4 2431 15-Dec-87 18:19:03 012 ?TRAP-I-Clock reset 4 1080 15-Dec-87 11 flaps (so far) Dave
karn@faline.bellcore.com (Phil R. Karn) (12/17/87)
The solution to the problems Dave describes seems awfully obvious: JUNK X.25!! Valuable computer networking people like Dave Mills have enough important problems to solve without wasting time on artificial ones like limits on the number of virtual circuits a network interface will support. When we (Bellcore) came up on the ARPANET, we had a choice between X.25 and 1822/HDH. I chose the latter and I've never had cause to regret it. If people had X.25 DTEs that they REALLY have to have communicate through the ARPANET, this should have been done with separate end-to-end boxes that encapsulate X.25 inside 1822 messages. The decision to inflict X.25 on the rest of us innocent Internetters is still utterly incomprehensible to me. Phil
oconnor@SCCGATE.SCC.COM (Michael J. O'Connor) (12/17/87)
I second the motion!
tom@tsdiag.UUCP (12/23/87)
Phil, you of all people should not confuse an implementation problem with a protocol problem, Dave CLEARLY states that the problem is the atrifical limit on the number of VC's SUPPORTED BY THE ACC 5250 interface and driver. The interface (ACC 5250) should be trashed (Dave suggests re-vamping it). so i think your statement should correctly read: JUNK ACC 5250! not x.25 as you all to well know I prefer the basic assumptions of x.25 over the ones of a datagramme based protocol. but we agree that we disagree! -- Thomas A. Moulton, W2VY Life is too short to be mad about things. Home: (201) 779-W2VY Packet: w2vy@kd6th Voice: 145.190 (r) Work: (201) 492-4880 x3226 FAX: (201) 493-9167 Concurrent Computer Corp. uucp: ...!ihnp4!hotps!ka2qhd!w2vy