On Fri, 21 Sep 2007, der Mouse wrote: >> That's the main question, are you really helping people and the >> Internet globally by letting people from the other side of the planet >> connect to your server to get time. > > I don't know, but I think so. > >> Wouldn't the Internet be better off if everybody was[sic] connecting >> to a server close to him. > > (There's this thing called a "question mark", which is conventionally > used in place of "." at the end of a sentence that's phrased as a > question. You might want to look into it; using it correctly increases > ease of reading the resulting text, and, in some cases, disambiguates > between multiple possible meanings.) > > Would it? Perhaps. If you can give me a way I can work towards that > end for approximately the same resource investment that my NTP Pool > membership is costing me, I'll be interested. Until then, I'll > continue my pool membership, on the basis that the world with me in the > pool is (slightly) better than the world without it. > >> Ideally, we should have a network divided into independent pyramids >> (stratum 1s on top). > > I strongly disagree. There should be cross-links at every level, to > reduce the damage from falsetickers, low-stratum falsetickers. (They > shouldn't exist, but they do. Stuff breaks.) Ideally the cross-links > should be between toplogically close machines.... > >> Right now, there is all kinds of cross-links between the pyramids and >> that's why I say ntp is a mess and I have come to ask myself if the >> pool was really helping things out. > > The redundancy the cross-links introduce is essential to detecting and > ignoring falsetickers. Otherwise, there'd never be any need to get > time from more than one place, and NTP *would* be pyramid-structured. > > Would I get worse time if I synced to a machine on the other side of > the globe rather than a local machine? Of course...other things being > equal. Other things are not always equal, though, and I'd much rather > get high-jitter time from a machine in India than time off by half an > hour from a machine at my own ISP. Checking against distant machines > provides a sanity check on the local machines, and NTP's clock > selection algorithms will normally prefer the machines that are giving > me the best time - which usually means the closest ones - anyway, using > the others for sanity-checking, and fallback when the sanity-checking > rejects the close machine(s). This seems to me to give the best of > both worlds.
Listen to this, your ISP would run 3 ntpservers, each one at the bottom of an independent pyramid. Each one of the 3 servers could even be at the bottom of 3 independant pyramids for a total of 9 independent pyramids. So there you go, that's all you need to prevent against false tickers. No need for a whole bunch of messy cross-links between the pyramids in my humble opinion. I have been designing failsafe network communication architecture for very large projects and beleive me, ntp is a big mess right now. -Lou > > /~\ The ASCII der Mouse > \ / Ribbon Campaign > X Against HTML [EMAIL PROTECTED] > / \ Email! 7D C8 61 52 5D E7 2D 39 4E F1 31 3E E8 B3 27 4B > _______________________________________________ > timekeepers mailing list > [email protected] > https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers > Louis http://blogtech.oc9.com _______________________________________________ timekeepers mailing list [email protected] https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
