Christopher-Vance@adfa.oz.au (Christopher JS Vance) (05/17/91)
Situation: We have a campus LAN with a Convex, half a dozen Pyramids, two dozen Apollos, three or four dozen Suns, and a gaggle of lesser machines. We run ntp on one of the Suns, have a timed master on the same machine and run timed slaves on the Convex, Pyramids, and Apollos. The timed master is compiled with MNOADJUST defined to stop ntp and timed messing each other up (it was a mess till I found the sources we currently use). Question: Is there some problem with Sun hardware (we have 3's and 4's) which would stop Sun shipping either ntp or timed? Why do all the other machines come with timed? Do Sun not consider synchrony important? Recommendation requested: Should we run a mixed ntp/timed network (as at present)? Should we convert to using ntp throughout? Should we use some other mechanism to keep our machines in synchrony? What overheads are there in running ntp? Timed? Do you have a better suggestion? -- Christopher
bob@MORNINGSTAR.COM (Bob Sutterfield) (05/17/91)
We use xntp exclusively in a setup similar to yours. One SPARCstation running PPP is our IP gateway to the world, and runs an xntpd that chimes with several extralocal clocks. All the other machines on our local network run xntpd or ntpd and chime with each other and the gateway machine. The load on the systems and the network is somewhere between unmeasureable and miniscule. If a machine had previously run timed when delivered, that line is commented out of its rc scripts and replaced with an NTP daemon invocation as soon as I get it compiled and running on each new architecture. In your situation, your strategy of running a harmless timed master is probably a good idea, to help with those machines that can't yet run xntpd. I don't know why Sun doesn't provide some sort of timekeeping system, especially considering how confused applications get when NFS clients and servers have different ideas about the time. A few vendors, but not many, are now shipping timekeepers with their systems. Perhaps it's a matter of waiting until the "research toys" filter down into the "real world" of commercial demands.
billd@fps.com (Bill Davidson) (05/18/91)
In article <9105171342.AA23309@volitans.morningstar.com> bob@morningstar.com (Bob Sutterfield) writes: >I don't know why Sun doesn't provide some sort of timekeeping system, >especially considering how confused applications get when NFS clients >and servers have different ideas about the time. A few vendors, but >not many, are now shipping timekeepers with their systems. Perhaps >it's a matter of waiting until the "research toys" filter down into >the "real world" of commercial demands. I've always thought it strange too. I have a man page for timed on SunOS 3.4 but I can't find an executable. Sun has always had a /usr/include/protocols/timed.h though it's years out of date. The source for timed from 4.3BSD does compile and run fine on SunOS 4.1 provided that you update /usr/include/protocols/timed.h. I had troubles trying that with 3.4. I didn't try 4.0.3. My company ships timed with the FPS Model 500 series and when we were Celerity we shipped it with the 1200 series. I believe a number of other vendors (most notably Berkeley) do so as well. I don't know what Sun's problem with it was. Admittedly, timed is not great at keeping the time correct but at least it will keep everyone on your local net in agreement about which second it is. Timed is so simple and small; I see no reason not to include it. While NTP keeps better time, it's a lot bigger and more complicated and in more of a state of flux. Unlike timed it requires that the admin have an idea what is going on to set it up. If you are willing to put in a little time to figure out how set it up, you'll be a lot happier with it than you would be with timed; especially if you are on the Internet. Even if you aren't on the Internet and don't have a reference clock, xntpd's drift compensation is reason enough alone to use it instead of timed. Still, a lot of people don't want to know about it; they just want things to work right. --Bill (I am NOT an offical spokesperson for FPS Computing Inc.) Davidson
bob@MorningStar.Com (Bob Sutterfield) (05/20/91)
In article <17835@celit.fps.com> billd@fps.com (Bill Davidson) writes: In article <9105171342.AA23309@volitans.morningstar.com> bob@morningstar.com (Bob Sutterfield) writes: I don't know why Sun doesn't provide some sort of timekeeping system... Perhaps it's a matter of waiting until the "research toys" filter down into the "real world" of commercial demands. I've always thought it strange too. I'm informed by a reliable source within Sun that From: DeepThroat@Somewhere.Sun.COM (Deep Throat) To: bob@morningstar.com Subject: Re: synchronizing a LAN-full of machines Sun will deliver SVR4 with NTP.. A V3 daemon if [it gets] done in time. So, as I speculated, there's no sinister plot against timekeeping, and no negligence, merely a matter of waiting until they get around to it.