Help to measure the Fiasco.OC's real-time capabilities

[Gernot Heiser]

On 23 Mar 2015, at 11:13 , Reinier Millo Sánchez <rmillo at uclv.cu> wrote:
> 
> I want to use Fiasco.OC to construct real-time's systems. I want to do some benchmarks over Fiasco.OC, to measure the performance in different hardware platforms and compare the results with tests made to others solutions based maybe on Linux Kernel.
> 
> For now i want to test the real-time capabilities on some platforms, measuring some aspects of Fiasco.OC (in time or cycles):
> - Execution mode switch (user-kernel)
> - Thread/Context switch on the CPU
> - Scheduler algorithm
> - IPC Mechanism
> - Thread/Task creation/destruction
> 
> Somebody have measured this elements in Fiasco.OC or can give me some ideas?

Measuring real-time performance is somewhere between hard and impossible.

The problem is that in order to make any real-time guarantees, you need a safe upper bound on the worst-case latencies. You can’t get those with measurements, you’re likely to under-estimate the WCET by orders of magnitude.

The only sound WCET analysis of a protected-mode OS I’m aware of is the one done on seL4, see
- http://www.ssrg.nicta.com.au/publications/nictaabstracts/Blackham_SCRH_11.abstract.pml
- http://www.ssrg.nicta.com.au/publications/nictaabstracts/Blackham_SH_12.abstract.pml

Gernot