[LITMUS^RT] A masters thesis that uses LitmusRT

[Björn Brandenburg]

On 01 May 2014, at 19:55, Glenn Elliott <gelliott at cs.unc.edu> wrote:

> 
> On May 1, 2014, at 1:53 PM, Björn Brandenburg <bbb at mpi-sws.org> wrote:
> 
>> 
>> On 01 May 2014, at 19:45, Glenn Elliott <gelliott at cs.unc.edu> wrote:
>> 
>>> Google Scholar notified me of this recent masters thesis: “Global EDF Scheduling for Parallel Real-Time Tasks” by Jing Li from Washington University in St. Louis.  The last few chapters compares the performance of Litmus-aware OpenMP running on Litmus's G-EDF vs. a fixed-priority OpenMP package.  Figure 9.3 shows how each OpenMP thread coordinates with Litmus.
>>> 
>>> http://cse.wustl.edu/Research/Lists/Technical%20Reports/Attachments/1046/Jing_Li_MS_Thesis%201.pdf
>>> 
>> 
>> Interesting, thanks for sharing! Concerning Figure 9.3, I’m surprised this worked at all. I guess we can add OpenMP compatibility to our feature list now…  ;-)
>> 
>> - Björn
> 
> Compatible at least for this configuration.  I worry that things could go bonkers as soon as budget enforcement is enabled.  I think mayhem would ensue if the job sequence numbers got out of sync among the parallel threads.

Yes, we should perhaps decouple the concepts of “how many times has sleep_next_period() been called” and “how many times have refreshed the budget”. 

We generally need a clearer separation between reservations and tasks. This was supposed to be part of Jonathan's Great Containers Rewrite, but maybe we can start with small things like confusing userspace when budgets are enforced..

- Björn