Edge Constrained Evaluation Results
To validate the edge deployability of MobilityNebula, we test our implementation under strict container resource caps in Docker.
We implement the nine queries (for supported queries per system) in:
- TStream (SpatialFlink): https://github.com/marianaGarcez/SpatialFlink
- GeoEKuiper: https://github.com/marianaGarcez/GeoEkuiper
- MobyDick: https://github.com/marianaGarcez/mobydick
All systems are evaluated with the same TCP stream input workload of 20k events/s. During these tests, we measure CPU usage, memory usage, throughput, and latency. Each query runs for 30 seconds.
We tested multiple configurations on a single-node setup under three memory caps: 512 MB, 2 GB, and 8 GB of RAM. We utilized:
- 4 vCPU + 8 GB RAM to simulate a Raspberry Pi 5
- 2 vCPU + 8 GB RAM to simulate an nROK device
Furthermore, we evaluated the conditions under which queries were either terminated or timed out without producing any sink output.
