yergeau@ee-cf.stanford.edu (Dan Yergeau) (12/13/90)
Are there any know incompatiblities between xntp and Ultrix's stock ntp? If so, are there any work-arounds? We can sync to other non-DEC equipment off campus, but all of our campus chimers are DEC. We can sync the DEC hardware from xntp, but xntp doesn't seem to be able to sync from the DEC. xntpdc requests just (e.g. to get a peer list) time out when directed at an Ultrix machine. Any advice would be helpful. Dan Yergeau yergeau@ee-cf.Stanford.EDU
ellozy@FARBER.HARVARD.EDU (Mohamed Ellozy) (12/13/90)
DEC runs ntpd (the trantor.umd.edu original code) while your other chimers run xntpd. There are minor interoperability problems. The worse is with the query programs, ntpdc does not speak to xntpd, and ntpq does not speak to ntpd (actually, they speak but get no answers). To get an xntpd host to sync to an ntpd host, you must put in the ntp.conf file a line like: peer 123.456.789.012 version 1 Of course, you can run xntpd on your DEC hosts, though it makes sense to run what you get. A long delayed question: How do people feel about ntpd vs xntpd? I went over to xntpd on the assumption that the latest version is by definition the best, but worry that the anti clock hopping code makes it too difficult to junk bad clocks. Mohamed
Mills@udel.edu (12/14/90)
Dan, DEC elected to use ntp, rather than xntp. The spytools of the one do not work on the other; however, the protocol itself interoperates just fine. If you need to spy on ntpd hosts, use the ntpdc program that comes with ntp.3.4. If you need to spy on xntpd and/or fuzzballs use the ntpq program that comes with xntp. Yes, I have recommmended to both DEC and Sun that they use xntp, but they get to choose themselves after all. Dave
Mills@udel.edu (12/14/90)
Mohamed, As mentioned previously, I have recommended xntpd for new installations for several reasons. First, the implementation will result in more accurate time, especially for those of us nuts no trying to split the millisecond. Second, its reference-clock implementation is much more solid than ntp. Third, it includes cryptosum authentication features, which when you end up winding the springs of 5000+ hosts, really is a comforting thing. Fourth, its spytools are compatible with the spec and at least one other impelementation (fuzzballs). It has, of course, more codemass and requires more silicon to run and heats the machine a few more watts, which may be the reason DEC and Sun have resisted adopting it. I have not observed that xntpd is more likely than not to mischime. than ntp, but may I disingeniously suggest a version-3 implementation should put all those fears to rest. Another mission is to convert the spytools to SNMP; however, I fear agreement on a MIB would take at least a year. Dave