Xref: utzoo news.sysadmin:2548 comp.dcom.modems:4116 comp.unix.questions:14797 Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!cbmvax!grr From: grr@cbmvax.UUCP (George Robbins) Newsgroups: news.sysadmin,comp.dcom.modems,comp.unix.questions Subject: Re: How to speed up uucp with Telebits(only getting 800 chars/sec ) Message-ID: <7243@cbmvax.UUCP> Date: 9 Jul 89 20:15:02 GMT References: <200038@hrc.UUCP> <1160@intercon.UUCP> <1989Jul8.042330.19789@wobble.uucp> <7102@swlabs.UUCP> Reply-To: grr@cbmvax.UUCP (George Robbins) Organization: Commodore Technology, West Chester, PA Lines: 178 In article <7102@swlabs.UUCP> jack@swlabs.UUCP (Jack Bonn) writes: > In article <1989Jul8.042330.19789@wobble.uucp> dlu@wobble.UUCP (Doug Urner) writes: > > The best I have been able to do between > >wobble and uunet (both Trailblazer+'s and both running the link to the modem > >at 19.2) is in the same range. .... > > but to uunet better than 900 is a real surprise. Any thoughts? ... > I think the problem is with uunet. Both pcrat (Rick Richardson) and I are > getting reduced throughputs to uunet. Let me look: > > System Mode Bytes Ave. CPS Burst CPS > uunet Rcvd 3430 1706 1776 > uunet Sent 1006756 690 982 > > This is on their (Non ATT) 800 line. Looks like uunet doesn't have > enough horsepower here. What are you other folks getting for throughput? Well, it's hard to make good estimates about performance without having some sense of the activity level at the far end. I seem 1200 cps type rates when my system (a 785) is unloaded and presumably uunet is lightly loaded. As soon as a news unbatching or batching run starts up on my machine, the speed drops by as much as 50%. The long term averages are pretty consistant though. I've attached some statistics for uunet and another system 'bpa' for comparison. The report is from an awk script that analyzes my SYSLOG file. Transfers are dropped in bins depending on the length of the transfer and apparent speed. Transfers of < 256 bytes are assumed to be itty bitty control files where the overhead is so large in comparision to the transmision time they are dumped in their own bin. Transfers with an effective speed of less thant 2500 bps are assumed to have taken place over 2400 BPS modems, which is a more or less supportable assumption. In examining these statistics, it is important to understand that the 'transmit' times reported by uucp with a Trailblazer tend to be completely bogus! The trailblazer buffers several (4->32k?) of data and claims it is 'done' before this data has actually been transmitted. The effect is that the first [buffer size] of any transfer can go as fast as 19.2KBPS and since most uucp transfers tend to be pretty small files, the tranmit times are badly skewed by this reporting retry. For a concrete expample. I recently changed the news batch size for bpa from 50K-byte to 250K-byte. Prior to this change, transmit rate averaged ~33% faster than receive, afterwards the disparity was chopped by a factor of two. What does this mean in terms of uunet/trailblazer performance? As the administrator of a fairly large multi-user unix system, I sense maybe half the "slowdown" is on my end. As long the overall transfer rate is about 4 times what I'd get on an olde fashioned 2400 baud modem, I'm fairly content. Someone running a lightly loaded hotshot 386 system might feel that sluggish performance on the part of uunet is costing them some bucks. The question of whether trailblazers are really 'faster' in the long run than a 'constant speed' V.32 modem remains an open issue. By the way, we dial directly into uunet over telco/AT&T lines. Anybody using various economy connections or dialing out thru a PBX might want to try it straight and see what effect this has. It would nice if Rick would publish some statistics on his end. We've shoved > 100M-Byte/month from/to uunet for the last 3 months and are only one of many sites. Another report I have indicates we move 14 M-byte/day over an average of .41 active transfers at any given time. I suspect Rick's numbers might be a teensy bit larger. uunet ================================================================================ system transfer files fail M-bytes hours speed peak retry ======= ============ ====== ==== ======== ====== ===== ===== ===== March 1989: uunet fast receive 1768 0 23.678 9:27 6959 bps 13334 bps fast send 697 0 5.984 1:31 10941 bps 15371 bps 0.2% slow receive 958 0 11.510 16:10 1976 bps 2444 bps slow send 252 0 3.206 4:45 1869 bps 2466 bps 0.2% misc control 3664 0 0.375 1:09 April 1989: uunet fast receive 3615 0 98.890 36:00 7630 bps 13277 bps fast send 724 0 6.426 1:43 10339 bps 16840 bps 0.1% slow receive 227 0 1.931 3:19 1613 bps 2500 bps slow send 15 0 0.080 0:06 1979 bps 2376 bps misc control 4794 0 0.506 1:45 May 1989: uunet fast receive 3703 0 132.472 48:49 7535 bps 13508 bps fast send 791 0 7.189 1:53 10600 bps 16987 bps 0.1% slow receive 166 0 0.949 1:39 1587 bps 2457 bps slow send 4 0 0.005 0:01 869 bps misc control 4769 0 0.463 1:50 June 1989: uunet fast receive 3312 0 109.088 42:06 7195 bps 13792 bps fast send 560 0 4.111 1:02 10889 bps 16568 bps 0.3% slow receive 154 0 1.257 1:53 1844 bps 2494 bps slow send 6 0 0.012 0:01 1329 bps 1165 bps misc control 4167 0 0.396 1:41 July 1989: uunet fast receive 741 0 26.443 9:18 7888 bps 13664 bps fast send 117 0 0.954 0:12 12565 bps 16017 bps slow receive 25 0 0.263 0:22 1966 bps 2402 bps misc control 953 0 0.102 0:21 bpa ================================================================================ system transfer files fail M-bytes hours speed peak retry ======= ============ ====== ==== ======== ====== ===== ===== ===== January 1989: bpa fast receive 5982 0 73.066 24:15 8368 bps 13437 bps fast send 550 0 12.014 2:44 12192 bps 17186 bps slow receive 190 0 2.297 3:05 2067 bps 2499 bps slow send 22 0 0.454 0:34 2191 bps 2263 bps misc control 6764 0 0.868 2:40 February 1989: bpa fast receive 5902 0 73.180 25:17 8037 bps 13042 bps fast send 518 0 12.190 2:55 11587 bps 17260 bps slow receive 206 0 2.527 3:17 2137 bps 2490 bps slow send 30 0 0.544 0:41 2195 bps 2270 bps misc control 6666 0 0.854 2:40 March 1989: bpa fast receive 5509 0 70.681 28:25 6905 bps 13103 bps fast send 1246 0 36.633 9:35 10609 bps 17076 bps 0.1% slow receive 890 0 11.174 14:33 2131 bps 2495 bps slow send 547 0 15.830 20:13 2174 bps 2301 bps 0.1% misc control 8210 0 0.966 2:58 April 1989: bpa fast receive 6120 0 74.420 28:28 7259 bps 13208 bps fast send 2549 0 73.646 19:03 10730 bps 16847 bps 0.1% slow receive 241 0 2.783 3:46 2047 bps 2458 bps slow send 245 0 6.993 8:54 2179 bps 2211 bps misc control 9175 0 1.015 3:01 May 1989: bpa fast receive 6061 0 74.971 29:25 7076 bps 13252 bps fast send 2633 0 76.044 19:43 10704 bps 17543 bps 0.1% slow receive 747 0 9.115 12:21 2049 bps 2499 bps slow send 594 0 17.043 21:35 2192 bps 2461 bps 0.1% misc control 10047 0 1.096 3:11 June 1989: bpa fast receive 4561 0 57.710 23:36 6788 bps 13206 bps fast send 1787 0 52.122 13:48 10490 bps 17553 bps 0.1% slow receive 1383 0 16.984 22:29 2097 bps 2489 bps slow send 1221 0 35.541 44:40 2210 bps 2486 bps 0.1% misc control 8960 0 0.964 2:50 July 1989: bpa fast receive 1208 0 14.708 5:37 7253 bps 12603 bps fast send 140 0 10.263 3:14 8817 bps 17288 bps 0.1% slow receive 255 0 3.292 4:19 2110 bps 2465 bps slow send 75 0 6.941 8:48 2187 bps 2388 bps misc control 1680 0 0.208 0:40 -- George Robbins - now working for, uucp: {uunet|pyramid|rutgers}!cbmvax!grr but no way officially representing arpa: cbmvax!grr@uunet.uu.net Commodore, Engineering Department fone: 215-431-9255 (only by moonlite)