Resource Isolation Demo

A demo that shows the timing-accuracy difference between a TIMPANI-scheduled workload and a normal periodic (sleep) workload under CPU load, using physical LEDs.

• Normal workload: keeps a 500ms period with sleep(0.5s) -> the period drifts once CPU load rises
• TIMPANI workload: activated on a 500ms period by the TIMPANI scheduler -> keeps an accurate period regardless of load

See Introduction.md for background and design intent.

How It Works

[Joystick button]  ->  serial-bridge  ->  KUKSA Databroker  ->  kuksa-bridge  ->  stress-ng ON/OFF (toggle CPU load)
[Rotary turn]      ->  serial-bridge  ->  Pullpiri          ->  workload containers Launch / Terminate
[TIMPANI signal]   ->  led-timpani-controller  ->  LED (accurate 500ms)
[Normal sleep]     ->  led-normal-controller   ->  LED (delayed under load)

There is a single key signal: Vehicle.Cabin.ResourceIsolation.ButtonPressed (written by serial-bridge, read by kuksa-bridge).

Layout

Two deployment modes are provided.

Mode	Description	Location
single_node	Everything runs on one PC (no SSH)	single_node/
multi_node	Split across master / guest nodes	multi_node/

Each mode shares the same sub-folder structure. The multi_node structure is split into nuc_master and nuc_guest, each containing the sub-folders that belong to that node. The Arduino sketches used by both modes are kept once in common/arduino.

<mode>/
├── arduino/           # compile / install scripts and udev rules
├── serial-bridge/     # serial <-> KUKSA / Pullpiri bridge, VSS spec, workload YAML
├── kuksa-bridge/      # subscribes to the button signal -> controls stress-ng
├── led-*-controller/  # LED control containers (timpani / normal)
├── monitoring/        # Prometheus + Grafana
└── test_script/       # run automation scripts *

Usage

Both start and stop go through test_script. Configuration is done by editing only the USER SETTINGS section at the top of test_script/config.env.

Single Node

Runs on one PC. See single_node/test_script/README.md for details.

cd single_node/test_script

./preflight/prepare.sh              # 1) one-time: build images + upload Arduino
./run_resource_isolation_flow.sh    # 2) start
./stop_resource_isolation_flow.sh   # 3) stop

Multi Node

Run preflight on each node first, then run the flow from the master PC. See multi_node/test_script/README.md for details.

# one-time on each node
./preflight/prepare_master.sh       # on the master node
./preflight/prepare_guest.sh        # on the guest node

# on the MASTER PC
cd multi_node/test_script
./run_resource_isolation_flow.sh    # start (orchestrates both nodes over SSH)
./stop_resource_isolation_flow.sh   # stop

Testing

After running run_resource_isolation_flow.sh:

1. Turn the rotary to 120 and press its button -> workload containers are launched. (once the flow detects the databroker log, it proceeds automatically)
2. Press the joystick button -> toggles the CPU load (stress-ng).
3. Grafana dashboard (http://<node-ip>:3000) -> compare the two LED periods.
   • TIMPANI LED: stays ~500ms regardless of load
   • Normal LED: period grows beyond 500ms under load

Notes

• Detailed values such as paths and image versions can be overridden in the ADVANCED SETTINGS section of config.env.