Hi Björn, On 21-09-07 08:22, Björn Gabrielsson wrote: > On Thu, September 20, 2007 21:00, Arnold Schekkerman said: >> My theory: a st1 may fluctuate due to load, temperature changes, signal
Yeah, you are right, temperature changes are no valid argument for a reference clock. It wouldn't be much of reference then. Thanks for pointing it out ;-) >> loss, etc. which is filtered by a st2. A st2 is less popular, so a well >> configured st2 only fluctuates due to temperature changes. Therefore, it's >> best to have a local time source or sync with st2 :-) > > S1 servers with good timesource will not be filtered in any good way by a > S2 server. A GPS timesource will give a PPS-pulse at better than 1us. This > is some decades better than achivable over a LAN - more so a WAN. Signal Yes, having your own ref clock is best for *absolute accuracy*. However, when _using_ an S1 over the net, you lose that aspect anyway. Another aspect one can consider is *continuity*. False tickers exist and if a S1 server votes out its own ref clock for some reason, it will jump. That's why you need at least 4 servers. If one S1 becomes bad _and_ the S2 has not so many S1 sources, then it will (most likely) be influenced by the bad S1 (filtered effect). An S3 server with many S2's will run very smooth, no matter what happens upstream (unless its own connection is the problem of course). These two aspects basically brought me to my conclusion: have a local ref clock (best abs. accuracy), or be down in the tree with many sources (smooth, continuous time). As I won't have good absolute accuracy on my asymmetric ADSL anyway, my thoughts are mainly based on smooth continuous time with reasonable accuracy. So, I carefully selected S2 and S3 servers with different upstream S1's. > [deleted correct explanation why I was wrong with ref clock inaccuracy] > Find a S1 without congested network pipes. That will give the "best" > performance! But maybe you do not need "best" performance on many > computers. AND if you need the performance its not hard or very expensive > to setup your local GPS time source these days. Because (official?) S1 servers could not handle the increased load, the pool exists in the first place. That's also why I assume an S2 has not configured many S1's. BTW When my EUR 10,- electronics arrive, my S3 server will become S1 using the DCF 1 PPS radio signal. Not that I need that accuracy (or continuity) but just for the fun of it :-) Arnold _______________________________________________ timekeepers mailing list [email protected] https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
