Commit 73264cda by FritzFlorian

Add notes on our current measurement setup.

parent 0b4580ee
......@@ -4,6 +4,59 @@ A collection of stuff that we noticed during development.
Useful later on two write a project report and to go back
in time to find out why certain decisions where made.
## 24.06.2019 - Further Benchmark Settings and First Results
As a further option to reduce potential inference of our benchmark
process we add the `CPUAffinity` option to systemd, setting the
CPU affinity of all processes started by the startup procedure
to core 0, thus isolating cores 1 to 7 for our benchmarks.
For docs on this see [the systemd manual](https://www.freedesktop.org/software/systemd/man/systemd.exec.html).
(Please note that this is not prefect, it seems that a kernel parameter
would perform even better, but our current measurements show this
working well enough to not go that deep a long as we do not see the
need in further testing).
We could potentially better our results further by using a preemtive
patch on a newer kernel, e.g. Arch linux with preemt patch.
Our first results look promising, and show that our measurement
methodology can indeed work (measured on the banana PI with the
above settings). Backgroud tasks are started from core 1 to 7
with 500us work, 5000us pause. Benchmarks are started pinned
to core 1 to 7:
FFT without background tasks:
<img src="./media/2ad04ce5_fft_boxplot_isolated.png" width="400"/>
FFT with background tasks:
<img src="./media/2ad04ce5_fft_boxplot_multiprogrammed.png" width="400"/>
Matrix without background tasks:
<img src="./media/2ad04ce5_matrix_boxplot_isolated.png" width="400"/>
Matrix with background tasks:
<img src="./media/2ad04ce5_matrix_boxplot_multiprogrammed.png" width="400"/>
The plot's include all outliers, showing that the workst case times can
be measured and compared under these circumstances.
One option we might change in the future is how/when the background
preempting workers are executed, to better 'produce' bad situations
for the scheduler, e.g. have a higher active time for the workers
(close to maybe 1/4 of the benchmark times), to actually fully
remove one worker from a few tasks (we would like to do that with
our tests on different queue types, to e.g. see if the locknig queue
actualy is worse than lock free implementations by blocking up other
tasks).
## 14.06.2019 - Benchmark Settings/Real Time Execution on Linux
Our set goal of this project is predictable execution times.
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or sign in to comment