Pipeline Performance

End-to-end runtime and memory benchmarks for the city2tabula -extract-features command across three municipality-scale datasets.

---

Environment

Component	Version / Spec
CPU	Intel Core Ultra 9 185H (22 cores)
RAM	62 GB
OS	Ubuntu 24.04
PostgreSQL	18.4
PostGIS	3.6

All datasets were fully loaded into 3DCityDB before timing started. No import or preprocessing time is included in the reported runtimes.

---

How Numbers Were Collected

Wall time and Go process RSS — /usr/bin/time -v:

/usr/bin/time -v ./city2tabula -extract-features 2>&1 | tee run_<city>.log

PostgreSQL RSS — sampled every 500 ms while the pipeline ran:

(while true; do
    ps -o rss= -p $(pgrep -d, postgres) 2>/dev/null \
      | tr ',' '\n' | awk '{s+=$1} END {printf "%d\n", s/1024}';
    sleep 0.5;
done) > pg_mem_<city>.log &

Kill the monitor after the run:

kill $(jobs -p)

Surface count — after extraction completes:

SELECT COUNT(id) FROM city2tabula.lod2_surface_raw;

---

Results

Dataset	LoD	Buildings	Surface count	Wall time	Throughput	Peak PG RSS	PG idle baseline	PG increment
Freiburg (DE)	2	130,191	2,297,558	10m 28s (628 s)	3,657 surf/s	7.5 GB	1.6 GB	5.9 GB
Vienna (AT)	2.1	130,000	2,753,166	21m 33s (1,293 s)	2,129 surf/s	8.6 GB	2.3 GB	6.3 GB
Deggendorf (DE)	2	125,991	1,768,124	8m 11s (492 s)	3,596 surf/s	7.4 GB	1.6 GB	5.8 GB

---