Re: Notes from the systemtap BOF

[Mathieu Desnoyers <compudj at krystal dot dyndns dot org>]

* William Cohen (wcohen@redhat.com) wrote:
> My understanding is that LTT has some support to take into account the 
> possible skew between the process time stamp counters.
> 

Yes. LTTV, the next trace reader generation of LTT, interpolates from
microsecond timestamps and the 64 bits TSC value, which are both recorded at the
beginning and end of a buffer, a time value taking account of both the
synchronisation between the CPU and the high resolution TSC on each of them.
This is counted in nanoseconds. Doing periodic recalibration of timestamp VS TSC
does help getting rid of CPU clock drift.

We cannot guarantee a precision more accurate than the microsecond : we are in
fact limited by the external clock source. If we could get a nanosecond accurate
clock source, world would be perfect.

We already used a GPS as an high resolution clock source to record the frequency
drift of some CPU. There is a master thesis around about this subject, made by
Hicham Marouani under the direction of Michel Dagenais, from École Polytechinque
de Montréal.


Mathieu


OpenPGP public key:              http://krystal.dyndns.org:8080/key/compudj.gpg
Key fingerprint:     8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68