skl@van-bc.UUCP (Samuel Lam) (12/08/89)
In article <BOB.89Dec7131357@volitans.MorningStar.Com>, bob@MorningStar.Com (Bob Sutterfield) wrote: >NNTP is the wrong place to manage or communicate time if you intend to >use the time for something that matters to you. But it isn't really "absolute time" that the two NNTP peers are trying to communicate here, it's just a "magic token" that point backs to a point in the server's time. It's like some LISP systems which has a built-in "time travelling machine"; You could ask it to give you a "ticket" for the moment "now" at any time, and you keep that ticket and present it to the time machine if/when you want to go back to that instance in time. In the NNTP world, we just need a way to tell the server that we would like to come back later to this point in time to pick up where we left off, and ask for a ticket which will bring me back here later. What the absolute time of the moment is should only be relevent to the server, if anyone. Having said that, how about having a new NNTP command "TICKET" which may, depending on the implementation, returns the return value of time() in %ld format, and have the "NEWNEWS command extended to optionally accept "TICKET non-blanks" instead of "yymmdd hhmmss" as the go-back-to point? Having the legal value of a ticket being a series of non-blank characters will allows the server to decide how it wants to keep track of the tickets it issued. Of course, a magic ticket is only redeemable at the outlet that issued it originally. And the ticket could be of the form "...@server.domain.name" to ensure that. To the client, the ticket is just a string of characters, and it wouldn't need to know the format of it beyond that. In the ideal world where everyone's clocks are synchronized, this discussion would be irrelevent, but in the imperfect world we live in, we need realistic and practical solutions. ...Sam -- Samuel Lam <skl@wimsey.bc.ca> or {uunet,ubc-cs}!wimsey.bc.ca!skl
eastick@me.utoronto.ca (Doug Eastick) (12/08/89)
skl@van-bc.UUCP (Samuel Lam) writes: >In the ideal world where everyone's clocks are synchronized, this discussion >would be irrelevent, but in the imperfect world we live in, we need realistic >and practical solutions. If you can talk NNTP, why not talk NTP (Network Time Protocol) as well? You get sub-second accurracy which would be (I assume) more than you would need for NNTP. Might as well strive for the perfect world... >Samuel Lam <skl@wimsey.bc.ca> or {uunet,ubc-cs}!wimsey.bc.ca!skl -- Doug Eastick UUCP: uunet!utai!me!eastick Mechanical Engineering eastick@me.utoronto.ca
skl@van-bc.UUCP (Samuel Lam) (12/08/89)
In article <1989Dec7.212233.12005@me.toronto.edu>, eastick@me.utoronto.ca (Doug Eastick) wrote: >If you can talk NNTP, why not talk NTP (Network Time Protocol) as >well? Is NTP available for Xenix V boxes with SCO TCP/IP then? >Might as well strive for the perfect world... I would very much like to, but maintainence resources here (and at some other NNTP sites I presume) is at a premium. I would rather maintain one slightly-bigger-than-needed package here than having to maintain two packages that are "just the right size" just so that I could use the first one. Afterall, it's one project versus two. Not having NNTP depending on NTP will also minimize the amount of thread one has to pull to get NNTP installed at a new site, especially one that's not BSD/SunOS based. >Doug Eastick UUCP: uunet!utai!me!eastick >Mechanical Engineering eastick@me.utoronto.ca ...Sam -- Samuel Lam <skl@wimsey.bc.ca> or {uunet,ubc-cs}!wimsey.bc.ca!skl
skl@van-bc.UUCP (Samuel Lam) (12/08/89)
In article <1989Dec7.212233.12005@me.toronto.edu>, eastick@me.utoronto.ca (Doug Eastick) wrote: >If you can talk NNTP, why not talk NTP (Network Time Protocol) as well? But wouldn't the upstream sites my site talks to via NNTP have to have NTP running as well? (Since those sites are under different administrative control, I can't just tell them to run NTP.) -- Samuel Lam <skl@wimsey.bc.ca> or {uunet,ubc-cs}!wimsey.bc.ca!skl
bob@MorningStar.Com (Bob Sutterfield) (12/09/89)
In article <82@van-bc.UUCP> skl@van-bc.UUCP (Samuel Lam) writes: In article <1989Dec7.212233.12005@me.toronto.edu>, eastick@me.utoronto.ca (Doug Eastick) wrote: If you can talk NNTP, why not talk NTP (Network Time Protocol) as well? Is NTP available for Xenix V boxes with SCO TCP/IP then? One of the developers is inside HP, so SysV stuff is pretty well available. Ports exist to MS-DOS (and I believe VMS), so other hostile environments (meaning those without convenient BSD-like facilities) should be do-able. Please contact ntp@trantor.umd.edu for current details. Might as well strive for the perfect world... ...maintainence resources here is at a premium... it's one project versus two. NTP installation is fairly straightforward. Since it is a general facility, it provides benefits to activities other than news. For instance, time-dependent NFS weirdnesses decrease in frequency. Many sites, even those not running NNTP, consider correct timekeeping a benefit worth investing a day or so into.
lear@genbank.BIO.NET (Eliot Lear) (12/12/89)
I'll thank you for not bloating my binaries by reinventing wheels I already have installed in the form of NTP, simply because it's only one package. If time synchronization is so important to NNTP, then maybe Stan should distribute NTP with NNTP. It's a pretty easy package to install, assuming your OS supports adjtime and TCPIP. -- Eliot Lear [lear@net.bio.net]