Project - Noise Calibration
Mentor - Stanislav Shalunov
Weekly Update 9/10
Developer - Cesar Marcondes

Things Done (in the past weeks):
1) Created a linux-log_based patch for the log-based measurement
2) Created a linux-hash_based patch for the hash-based measurement
3) Created a linux-packet_based patch for the packet-based measurement that
keeps timestamping the packet as it goes through the network stack.
4) Extensive set of experiments under CPU 25%, 50%, 75%, and IO in 3 levels.

   Internal Queue Measurement (under CBR traffic and no coalescing) is pretty
constant with a long tail as the offered load increases, due to some context
switches to take care of CPU experiments. (see internal_queue.png
measurement).

    Does this clock skew graph (clock_skew.png) make sense?
    Basically under the packet-based measurement, I calculate the difference
between the packet been sent (dev_xmit) and been received (netif_rx) in linux.
It seems that there is no fixed time offset, but as # of packets (the x axis)
goes through there is a positive skew (perhaps one way delay - but the traffic
is CBR / or the clocks are really out of sync - strange since both machine are
identical) of about 300 microseconds in the 30 seconds experiment.

5) Created the version 2 of the final report (in the CVS) along with all the
measurements so far.

To be Done (in the last week):
6) Working on a queueing model based on Busy Periods M/G/1 analysis (gain
insight in the probabilistic model underneath).

=> If time permits, Windows modification - can I use one of the thrulay ports
for windows, of you guys?. This part got delayed due to UCLA policy of
disallow Windows Academia Alliance downloads (of Visual .NET and WindowsXP
activation license) if the student is not enrolled in the interquarter
(summer) - that it's my case.

Best regards,
Cesar Marcondes