[LITMUS^RT] feather-Trace Results

[Björn Brandenburg]

> On 31 Jan 2017, at 13:34, lambros bersentes <lambrosbersentes118 at hotmail.com> wrote:
> 
> I read the very first paper concerning the Litmus Rt framework (A Testbed for Empirically Comparing Real-Time Multiprocessor Schedulers) and i tried to reproduce the results on page 7 concerning the context switching costs and scheduler costs.

Cool, but just a note: that paper predates the public LITMUS^RT version and Feather-Trace, and used completely different measurement methods. 

> I used the feather-trace tool in order to measure the results on 3 task sets consisting of 10-20-50 same tasks. I am reading the stats.csv file which i get and to be accurate just the CXS and SCHED. I am getting all the results in cycles. My question is in order to be able to compare them with another tool (HSF)

Sorry, what is this ‘HSF’? Doesn’t ring a bell.

> which also gives statistics,however measured in seconds, is there any option to calculate all the statistics in seconds instead of cycles

Check out the --cycles-per-usec option of ft-compute-stats.

> or how can i output the current frequency of my CPU

On x86: cat /proc/cpuinfo | grep 'cpu MHz'

> on which the tests were done. The frequency of my CPU is not apparently stable as it changes depending on the task load.

That is a problem. Feather-Trace uses the cycle counter. If the frequency is not stable across the experiment, it will not be possible to convert the observed overheads from cycles into time units.

- Björn