[LITMUS^RT] Overheads

[Björn Brandenburg]

Hi Jorge,

> On 16. May 2019, at 22:45, Martinez Garcia Jorge Luis (PS-EC/ESB2) <JorgeLuis.MartinezGarcia at de.bosch.com> wrote:
> 
> I’m currently trying to compare scheduling overheads under different schedulers. To that end, I would like to understand the overhead name convention used in Litmus. After computing some statistics by means of ft-compute-stats, I can see the following categories:   CXS, RELEASE, RELEASE-LATENCY, SCHED2, SCHED, and SEND-RESCHED.
> Please correct me, if I am wrong, I assume that CXS accounts for the overhead related to process switching (context-switch overhead)

yes, this is correct.

> and SCHED + SCHED2 for process selection (scheduling overhead).  

Correct.

> Do RELEASE and RELEASE-LATENCY account for the execution of a release ISR (release overhead) and the delay until the release ISR starts execution (event latency)?

Yes: RELEASE is the cost in cycles of processing the release event in interrupt context, and RELEASE-LATENCY is the delta (in IIRC nanoseconds) between when the release should have occurred (i.e., the time at which the timer was programmed to expire) and when it was actually processed. 

> What does SEND-RESCHED account for?

The latency between when an IPI is sent and when it is processed by the remote core. 

> Once a job is completed there is some kind of overhead involved like the one related to the trap generated by its completion, does this overhead fall under one of those categories?

There is no extra event for job completions; any cycles spent on calling into the kernel to indicate job completion are part of the job’s regular execution cost. Dispatching the next job are regular scheduling and context-switch events. 

Best,
Björn

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 5061 bytes
Desc: not available
URL: <http://lists.litmus-rt.org/pipermail/litmus-dev/attachments/20190517/730868ee/attachment.bin>