On Wed, Oct 14, 2009 at 08:43:25PM -0700, Ask Bjjjrn Hansen wrote:
> On Oct 14, 2009, at 19:44, der Mouse wrote:
> >Lest anyone think otherwise, this is not hypothetical.
> 
> Graphs from carefully chosen[1] DNS servers in the pool.ntp.org  
> infrastructure shows a bit over twice as many queries over the minutes  
> around the top of the hour as any other similar time interval.  I'm  
> sure if the graph was granular down to seconds it'd be much more  
> extreme.

I've been collecting data as well and there are patterns everywhere.
The on-the-hour spike starts about a second before the hour with the
majority of clients hitting my server between 1.12 seconds and 3.31
seconds after the hour.  It slowly tails off till about 9 second when
it's reached normal levels.  At its highest point it's about 4 times
normal activity.

Looking further over the hour it's still interesting; there are spikes
every minute that are ~20% above normal with especially big spikes at 30
minutes (300% above normal and much more concentrated than the one on
the hour) and every 10 minute interval (150% above normal) and 5 minute
interval (80% above normal).  Note that that's above, so 300% above = 4
times.

All fun and seems to suggest that people are pretty crap at picking when
to request the time.  Then again, if we look at the amount of traffic
that actually occurs within these "busy" periods it's less than 5% of
everything else.  If I try and correct this 5% to take account of the
baseline activity that would be occurring if everybody was picking the
best times then I see 2% (depending on how I do it) of the overall
traffic in these spikes.

Even assuming a worst case of 5% I'd say that most people most of the
time (i.e. 95%) do a very good job of picking when to request time.


So, to those people who want to use a cron job.  I'd say go for it, just
make sure it's not too close to the hour, the minute isn't divisible by
5 and you put a random pause in (up to a second, as I see very reliable
spikes at 370ms to 400ms into each second) and you should be set.

-- 
  Sam  http://samason.me.uk/

p.s. plots available if people are interested
_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers

Reply via email to