root@odi.com (Operator) (09/08/89)
I've brought up the 4.3 timed under SunOS 4.0.3, more or less. The typical failure mode is that the master is alive but ineffective Tracing shows that it is getting an endless stream of acks from a particular slave with the same negative sequence number. When this is true, I can't turn tracing on. I get the feeling that only the master ever actually traces? Or is it true that if -t wasn't specified, then trace on is of no effect? I know this is not very coherent. In the trace I saw, of a master started -t, once the bad sequence numbers started to arrive then TRACEOFF requests also showed in the trace but the trace continued. Any direction as to what to investigate would be appreciated.
rusty@GARNET.BERKELEY.EDU (09/09/89)
Toss timed and use ntp. I tried the various known-to-work-on-suns patches for the 4.3 timed and was never able to get it to work well. ntp is much better anyways. You can get the ntp sources from trantor.umd.edu in the directory /pub/ntp.3.4. The file ntp-test.tar.Z is what everyone seems to be using. In the directory /archive is a collection of the mail that's been going around on the ntp mailing list. There is a bunch of documentation on ntp on louie.udel.edu in /pub/ntp. And I just noticed that the ntp-test.tar.Z file is in there as well so you don't need to go to trantor.umd.edu.
jpa@fps.com (Jeff Anderson) (09/11/89)
In article <8909081936.AA10167@garnet.berkeley.edu> rusty@GARNET.BERKELEY.EDU writes: >Toss timed and use ntp. I tried the various known-to-work-on-suns >patches for the 4.3 timed and was never able to get it to work well. >ntp is much better anyways. Well that's a responsible attitude. Come on guys, if it doesn't work on your machine, lets fix it. That's what other vendors have to do, and I can't believe it's that hard. We have enough standards already without implementing new ones because a single vendor doesn't like it. I have customers who use timed. I'm supposed to tell them to change? Because Sun's not happy? Timed is functional and it's an established standard -- maybe not the best, but it works TODAY on FPS machines, and DEC machines, and probably lots of others. Jeff Anderson jpa@fps.com FPS Computing arpa : ucsd!celerity!jpa@nosc 9692 Via Excelencia, San Diego, CA 92126 phone: (619) 271-9940
benson@odi.com (Benson I. Margulies) (09/13/89)
In article <617@celit.fps.com> jpa@fps.com (Jeff Anderson) writes: >In article <8909081936.AA10167@garnet.berkeley.edu> >rusty@GARNET.BERKELEY.EDU writes: >>Toss timed and use ntp. I tried the various known-to-work-on-suns >>patches for the 4.3 timed and was never able to get it to work well. >>ntp is much better anyways. > >Well that's a responsible attitude. Come on guys, if it doesn't work on >your machine, lets fix it. That's what other vendors have to do, and I >can't believe it's that hard. We have enough standards already without >implementing new ones because a single vendor doesn't like it. I have >customers who use timed. I'm supposed to tell them to change? Because >Sun's not happy? Timed is functional and it's an established standard -- >maybe not the best, but it works TODAY on FPS machines, and DEC machines, >and probably lots of others. A- You may be flaming the right vendor, but surely for the wrong reason. 1) rusty has nothing to do with Sun, so far as I know. 2) Sun don't even supply timed. Nor does it supply ntp. All they supply is tired old rdate, which is pretty inadequate for keeping a mare's nest of workstations in some faint sort of agreement. As Sun users, I (and I presume, perhaps incorrectly) rusty have better things to do with our time then port timed to SunOS. And our leverage with sun is the usual: nil. So don't flame us for looking for solutions wherever we can get them. B- NTP and timed are two different things. timed's excuse for existence is to take a bunch of equally (un)trustworthy clocks and try to keep them in sync. It sure seems to feel free about using a lot of CPU power to do so. NTP's purpose is to start with some source of reliable time (like a fuzzball or a WWV receiver) and distribute it hierarchically. Different sites fit the two different paradigms. It would be nice if both capabilities were integrated into one package. They ain't. -- Benson I. Margulies
smb@ulysses.homer.nj.att.com (Steven M. Bellovin) (09/14/89)
In article <617@celit.fps.com>, jpa@fps.com (Jeff Anderson) writes: > Timed is functional and it's an established standard -- Actually, NTP was just declared an elective Internet standard; newly documented in RFC 1119, hot off the "presses" (or at least disk drives) at NIC.DDN.MIL. Timed remains what it always was: some code developed at Berkeley, which others may or may not pick up.