[LITMUS^RT] experiment-scripts fork

[Glenn Elliott]

On Feb 24, 2014, at 8:54 AM, Björn Brandenburg <bbb at mpi-sws.org> wrote:

> 
> On 20 Feb 2014, at 20:46, Glenn Elliott <gelliott at cs.unc.edu> wrote:
> 
>> Since Jonathan has moved on to other things, I’ve decide to fork a copy of his experiment-scripts into the Litmus github account.  I guess I’ll take ownership for the time being.
> 
> Thanks for taking it over! I haven't worked with it yet, but I would welcome if it becomes the "official" experimental setup at some point. Is there anything major that needs work? (Also, is there are tutorial on the wiki?)
> 
> Manohar, you might want to add this repo to the VM image.
> 
> Thanks,
> Björn

There’s a basic tutorial in the github markdown README.  Before the fork, C-EDF experiments were broken due to CPU<->cluster mapping issues.  I’ve fixed this in the fork, but the fix is dependent upon the recent litmus/liblitmus cluster patches I proposed a few weeks ago.  This dependence wasn’t intentional—I just couldn’t think of an elegant workaround.

There are a few features on my todo list for the fork.  The scripts can parse sched_trace logs to get overhead measurements, etc.  I want to:
1) Compute response time information from sched_trace logs. (The scripts already do this for observed tardiness, so response time support is pretty trivial.)
2) Allow analysis to be constrained to a specified time window.  Statistical data is skewed for jobs that execute towards the end of an experiment, as tasks begin to drop out of the system (ex. average response time of jobs improves as there is less contention for the CPU).  Ideally, analysis should be limited to the window from synchronous release to the completion time of the last job of the first task to exit litmus.
3) A secondary method for generating schedulable task sets.  The scripts currently generate task sets with a set number of tasks.  This is useful, but we may want add a secondary generation method that leverages schedcat to generate _schedulable_ task sets.  I have a hacked up solution for my own experiments.  I’ll see if I can find the time to generalize them.

-Glenn