measure
  BMC best master clock
    computed from different infos including hop-count => measure latency between
      2 PTP master stations through a different number of routers
  use PTPd in "relay" mode

technical documentation
  calibrate/designate_the_accuracy of FITkit crystal using GPS signal
  take inspiration from
    http://ptpd.sourceforge.net/doc.html

notes (mostly done)
  NTP description + drastic configuration
  inturrupt handler each 1ms (== clock period == QNX "tick"), min == 10us
    The timeslice rate (for “round-robin” and “other” scheduling
    policies) is always four times the clock period (this isn't
    changeable).
  switch Cisco 3560
  traffic generator&analyzer Spirent SPT-2000A
  add thanks to Vaclav Simek (access to soldering station etc. at any time)
  Atom system was totally unusefull when ethernet load was 100%
    nothing worked and QNX always stopped working (freeze) after a short
    period of time while running demanding applications showed different errors
    then the QNX host with Atom started to fail almost in every load - after
    a few seconds of running compilation and the like the system
    inevitably hanged. Note that these problems appeared both with ethernet
    cable connected and disconnected
  after 18hours of frames with non-matching MAC addr 100% bandwidth load,
    the QNX C2D PC stopped to respond to ARPs
  weird behaviour of the network (maybe Cisco switch 3560?) which started to
    send frames not only to the traffic destination, but also to all other
    ports -> the network card on QNX stations had full bandwidth filled with
      frames not meant for them
    reason:
      maybe because of the high traffic, the ARPs stopped to "work" and switch
      decided to send frames everywhere
  QNX si neporadi se 100% load na interface => reboot needed
  cca after 10 minutes of the 100% traffic on the rt0 interface of NTP master computer,
    the connection stopped to work - reboot of QNX needed (hypoteza: mozna by stacilo
    restartovat proces obstaravajici sitovou kartu; hypoteza: sitova karta se dostala
    do nekonzistentniho stavu a proto QNX potom nekolikrat nemohl najet
    ani kdyz byl PC na chvili vytazeny ze zasuvky)
  nice - pro usera jine nez roota, ale mereni pod rootem, takze na max!
  u PTP zatizeni master => pruser
                 slave => temer bez rozdilu
  u NTP zatizeni master => temer bez rozdilu
                 slave => pruser (protoze tam ten vymenny mechanismus
                                  zprav zacina opacne)
  moznosti pokracovani BP
    performance of QNX messages/timers
    measure with relay/transparent_clock
    extend the measurements with measuring of timestamps (using Spirent etc.)
    discover the cause of performance problems on the Atom PC (why was NIC so red when compared to C2D PC where the NIC didn't show anything!)
    try adjusting the dispatch period of PTP/NTP messages and compare them
    figure out where the random delays from master in the long-lasting
      measurements come from (SW: RS232 driver, PCI driver
                              HW: many possibilities...)
    figure out exactly where the periodical subtle shifts in time on master
        come from
      integer rounding/truncation error (no!)
  simulace PTP/NTP
    http://www.eecis.udel.edu/~mills/ntp/html/ntpdsim.html
  zvetseninu pily v grafech + hypotezy

add to Alfresco
  http://www.eecis.udel.edu/~mills/ptp.html
  this todo
