As far as I understand, spikes are caused by SNMP clients that request time on the hour and half-hour. They might also be a ntpdate requests run from cron jobs. Finer structure of the peaks is very interesting. They have sharp front edge and exponentially decaying tail: http://leobodnar.com/files/100Mb-peak-NTP-pool.png This is probably a side-effect of switching/routing saturation and de-queueing.
>From what I have seen the spikes are defined by the infrastructure limitation >or NTP server packet handling capacity itself. They are like Dirac delta >functions - infinitely narrow and infinitely high. ntpd clients are usually very well behaved and spread much more evenly. Cheers, Leo > From: Joseph B <[email protected]> > To: [email protected] > Subject: Re: [Pool] Adding Stratum 2 servers in *some* underrepresented zones >> You can add 203.135.184.123 to China, it's a LeoNTP that coped with >> Snapchat and I've reconfigured my network layout so it won't melt this >> time (hence the new IP). http://leobodnar.com/LeoNTP/CCGS.php > > The graphing for the device is very cool, nice to have that granularity! > > I find the 40k-50k PPS spikes every few hours interesting. > > I wonder if it's specific to the AU zone or if it's more widespread? > > Cheers, > > Joseph _______________________________________________ pool mailing list [email protected] http://lists.ntp.org/listinfo/pool
