Ravinder.Chandhok@CS.CMU.EDU (Rob Chandhok) (10/12/90)
Well, go out of town for a week and what do you know comes up? Lots of mail about ntp... Regarding NTP on the Mac, I ported a particular ntp application from unix to the Mac as an MPW tool. "ntptime" syncronizes your Mac's clock via ntp, but doesn't do it continuously, like a unix ntp demon would. About using other time protocols (like UDP time and whatever) that ntp is much more robust and accurate. In fact, ntp is somewhat overkill for the 1/60th second resolution of the mac clock. I understand some AppleTalk based protocol should be used so MAc's don't have to depend on IP hosts for time information, but the fact of the matter is that there is an already existing extremely accurate hierarchy of time servers on the internet. Also, you could easily make a Mac into an ntp server, and just use DDP instead of UDP. Regarding the statement (from Pete Resnick?) that "the math is too hard for ntp", if anyone would like to implement a better form of ntp for the mac (cdev or init) then please take my sources! John Pane and I figured out how to convert from Mac time to ntp's notion of time. The rest of it is just getting and sending packets. ntp sources are copyright but usable for non-commercial use, that's how I got mine. The sources live on host gnome.cs.cmu.edu, in file /usr/chandhok/public/macntp/ntpsrc1.2.sit.hqx As mentioned before, do a "cd /usr/chandhok/public" after you log in as anonymous. You won't be able to connect to /usr/chandhok. All I want is for the copyright notices,etc., to be kept in the files. Maybe someday MacTCP will have ntp support also! So, let me know who is going to implement the next version :-) ! Rob
resnick@cogsci.uiuc.edu (Pete Resnick) (10/12/90)
Ravinder.Chandhok@CS.CMU.EDU (Rob Chandhok) writes: >About using other time protocols (like UDP time and whatever) that ntp is >much more robust and accurate. In fact, ntp is somewhat overkill for the >1/60th second resolution of the mac clock. You mean a mac does not need 200 picosecond precision?? :-) >I understand some AppleTalk based protocol should be used so MAc's don't >have to depend on IP hosts for time information, but the fact of the matter >is that there is an already existing extremely accurate hierarchy of time >servers on the internet. Also, you could easily make a Mac into an ntp >server, and just use DDP instead of UDP. Personally, I am trying to make my Mac MORE dependent on IP hosts. I mean heck, I got MacTCP for a reason. >Regarding the statement (from Pete Resnick?) that "the math is too hard for >ntp", if anyone would like to implement a better form of ntp for the mac >(cdev or init) then please take my sources! John Pane and I figured out how >to convert from Mac time to ntp's notion of time. The rest of it is just >getting and sending packets. Let me re-phrase: I took a look at the NTP RFC. The math is very complex and requires repeated packet sends and re-adjustment. My figuring was that I would not want to bog down my poor little SE with so many calculations and IP transmissions to set my clock when I don't need the 200 picosecond precision. Anywhere between 2 and 10 second precision was fine with me, so I used the UDP Time RFC. Wasting more than a couple of ticks just to get my clock set "reasonably" seemed absurd. pr -- Pete Resnick (...so what is a mojo, and why would one be rising?) Graduate assistant - Philosophy Department, Gregory Hall, UIUC System manager - Cognitive Science Group, Beckman Institute, UIUC Internet/ARPAnet/EDUnet : resnick@cogsci.uiuc.edu BITNET (if no other way) : FREE0285@UIUCVMD
Ravinder.Chandhok@CS.CMU.EDU (Rob Chandhok) (10/12/90)
>From: Pete Resnick <ux1.cso.uiuc.edu!resnick@iuvax.cs.indiana.edu> >Subject: Re: NTP on the Mac > >>Regarding the statement (from Pete Resnick?) that "the math is too hard for >>ntp", if anyone would like to implement a better form of ntp for the mac >>(cdev or init) then please take my sources! John Pane and I figured out how >>to convert from Mac time to ntp's notion of time. The rest of it is just >>getting and sending packets. > >Let me re-phrase: I took a look at the NTP RFC. The math is very complex >and requires repeated packet sends and re-adjustment. My figuring was >that I would not want to bog down my poor little SE with so many >calculations and IP transmissions to set my clock when I don't need the >200 picosecond precision. Anywhere between 2 and 10 second precision >was fine with me, so I used the UDP Time RFC. Wasting more than a couple >of ticks just to get my clock set "reasonably" seemed absurd. Let me re-phrase: the code that I ported to the mac already does all the packet sending & re-adjustment. It is not the ntp demon, so in about 5 seconds you could have an accurate time setting. IT seems to me that if you are going to bother setting the clock, that 2-10 seconds drift is a little loose. If you are going to rely on the time for using file servers, or other reasons for having a correctly synched clock, why not get it right? People waste more ticks with background screens... Oh well. It isn't really that big a deal, unless you really care abou thte time. We use the ntptime tool to ensure our timestamps are good on our projector files. Rob
resnick@cogsci.uiuc.edu (Pete Resnick) (10/12/90)
Ravinder.Chandhok@CS.CMU.EDU (Rob Chandhok) writes: >Oh well. It isn't really that big a deal, unless you really care abou thte >time. We use the ntptime tool to ensure our timestamps are good on our >projector files. Sorry. I didn't mean to make it that big a deal. I still plan to download the code. If I can make this real ntp without having the Mac having a fit, it would be much better than UDP time. But since 1.0a1 is not even out yet, this will wait for 1.1. Now, if I can just get back to my school work! pr -- Pete Resnick (...so what is a mojo, and why would one be rising?) Graduate assistant - Philosophy Department, Gregory Hall, UIUC System manager - Cognitive Science Group, Beckman Institute, UIUC Internet/ARPAnet/EDUnet : resnick@cogsci.uiuc.edu BITNET (if no other way) : FREE0285@UIUCVMD