November 2026 Summaries

Port Context Engine is a free maritime congestion-monitoring platform launched in October 2026 that uses public AIS vessel signals to show anchorage queues, berth arrivals, waiting times, and recent historical comparisons for Rotterdam, Antwerp-Bruges, Piraeus, and Hamburg. Developed by a single founder with AI-assisted coding, it processes roughly 860 AIS messages per minute through five Go services, NATS JetStream, and QuestDB running on a small VPS, while limiting coverage to ports where free terrestrial AIS data is sufficiently reliable. QuestDB supports the system’s time-based workload through ILP ingestion, PostgreSQL-wire reads, append-only vessel history, upserted port-call phases, and features including LATEST ON for current vessel state, ASOF JOIN for historical ETA context, and SAMPLE BY for minute-level downsampling. Raw positions remain in the database for seven days before a nightly archiver exports verified partitions to Parquet, while summarized positions and vessel records use automated retention policies. The project also documents operational lessons from QuestDB 8.2.3, including the need for independent safeguards before dropping data, table-rebuild procedures for port removal because row-level deletion is unavailable, connection recovery after database restarts, and schema choices that avoid ILP type-casting failures. Future work includes validating pilot-port results against external records, adding geospatial hashes, and moving ingestion to ILP over HTTP.

QuestDB presents a reproducible benchmark of replaying a full day of Databento EQUS.MINI U.S. equity top-of-book and trade data, comprising roughly 354 million book events and 7.5 million trades, to evaluate high-throughput, timestamp-ordered market-data streaming for backtesting. Using Apache Arrow batches over QuestDB’s QWP protocol, the tests streamed 57.6 GB of decoded data while keeping client memory relatively flat, with strict global ordering taking roughly 119–130 seconds at 3.5–3.8 Gb/s and parallel range-based replay reaching 14.2 Gb/s over a 12.5 Gbps network connection in 32.5 seconds. Localhost tests removed the network constraint and achieved up to 27.1 Gb/s with 16 connections, while a filtered NVDA replay took about 5.6–6.0 seconds in strict order. The results were obtained with data resident in page cache, an empty consumer function, and top-of-book-only rows, so disk access, deeper order books, consumer processing, hardware, network capacity, and schema width could materially affect performance. The accompanying code and commands are published for QuestDB open-source and Enterprise users, although Enterprise is required for the Databento connector used to load the benchmark data.