Test application for simulating a real-time periodic load
rt-app <config-file>
rt-app [options]-t<period>:<exec>[:policy[:CPUaffinity[:prio[:deadline]]]] -t ...
rt-app is a test application that starts multiple periodic threads in order to simulate a real-time periodic load. It supports SCHED_OTHER, SCHED_FIFO, SCHED_RR as well as the AQuoSA framework and SCHED_DEADLINE.
Code is currently maintained on GitHub: https://github.com/gbagnoli/rt-app
rt-app runs on GNU/Linux. It needs autoconf, automake, libtool and a recent compiler (mainly: gcc) for basic features. For advanced usage, the following optional libraries are needed: - qreslib for AQuoSA support (and, obviously a GENS_SCHED patched kernel) - a SCHED_DEADLINE patched kernel to support SCHED_DEADLINE policy. - json-c installed for reading config files. In case libjson is not available for your target, you can download and compile json-c from http://oss.metaparadigm.com/json-c/
rt-app <config_file> whereconfigfileisafull/relativepathtoajsonfile(lookunderdoc/for an example config file) or "-" (without quotes) to read JSON data from stdin.
The python merge script (/usr/share/doc/rt-app/examples/merge.py) can be used to combine json snippets into the config file: e.g: ./merge.py -o combined.json global.json resources.json thread0.json thread1.json thread2.json thread3.json
OR you can use commandline to define the taskset.
Keep in mind that on commandline it will never be possible to define resources and how tasks access them.
rt-app [options] -t <period>:<exec>[:cpu affinity[:policy[:deadline[:prio]]]]
br POLICY: f=SCHED_FIFO, r=SCHED_RR, o=SCHED_OTHER, q=AQuoSA br AFFINITY: comma-separated cpu index (starting from 0) i.e. 0,2,3 for first, third and fourth CPU
Note: when using AQuoSA scheduling, priority is used as percent increment for budget over exec time
-h, --help
show this help
-f, --fifo
set default policy for threads to SCHED_FIFO
-r, --rr
set default policy fior threads to SCHED_RR
-s, --spacing
msec to wait beetween thread starts
-l, --logdir
save logs to different directory
-b, --baselog
basename for logs (implies -l . if not set)
-G, --gnuplot
generate gnuplot script (needs -l)
-D, --duration
time (in seconds) before stopping threads
-K, --no-mlock
Do not lock pages in memory
-q, --qos
create AQuoSA reservation
-g, --frag
fragment for the reservation
--fifo,--rr,--qos
sets the default scheduling policy for all threads. --qos is available only when compiled with AQuoSA support enabled.
--logdir,--baselog
control log output (rt-app creates one log for each thread). If --logdir is not present, rt-app logs everything to stdout. When logging to file and -D (see below) is present, the file is written at the end of execution, during which all logging data is kept in memory.
--gnuplot
if it is present then rt-app outputs sample gnuplot scripts to $logdir (which must be passed too
--spacing,--duration
control the spacing between threads and their duration of execution, i.e. each thread is started every $spacing ms and the total runtime is $duration.
--frag (AQuoSA only)
fragments the reservation period by $fragmentation, i.e. reservation period is $frag times smaller than thread's period.
--no-mlock
do not lock threads' memory pages when using realtime scheduling classes
Each thread is specified by
--thread <period>:<exec>[:$POLICY[:deadline[:prio]]]:
period
: thread period in microseconds
exec
: thread WCET in microseconds
policy
: one of f=SCHED_FIFO, r=SCHED_RR, o=SCHED_OTHER, q=AQuoSA (if enabled)
affinity
: comma-separated cpu index (starting from 0) i.e. 0,2,3 for first, third and fourth CPU
prio
: thread priority in SCHED_FIFO/SCHED_RR/SCHED_OTHER, percentile increment for AQuoSA reservation budget w.r.t. WCET.
deadline
: deadline in microseconds (used ONLY for plotting!)