ATLANTIS

Network telemetry visualization platform for analyzing large-scale NetFlow data. SvelteKit frontend, SQLite backend, Rust ingestion pipeline.

ATLANTIS reads nfcapd captures from a NetFlow collector. The pipeline converts the captures into a SQLite database, and the dashboard visualizes that database.

Quick start

First, install the required tools — both Bun and Node.js are necessary, and nfcapd processing needs the nfdump build tools. You also need nfcapd capture files on disk.

1. Clone the project and build the tools.

   bash

   1	git clone https://github.com/flamboh/atlantis.git
   2	cd atlantis
   3	bun install
   4
   5	# A system nfdump installation does not work.
   6	git submodule update --init --recursive
   7	./vendor/scripts/compile-nfdump.sh

2. Describe your captures. Copy the templates, then edit them:

   bash

   1	cp .env.example .env
   2	cp datasets.json.example datasets.json

   In datasets.json, set root_path to your capture directory and name one source for each collector directory under it (dataset configuration explains the fields and the expected layout). In .env, set DEFAULT_DATASET to your dataset_id.

3. Build a database from one day of captures. Use your dataset_id and a date that has captures. The first run compiles the Rust pipeline and takes several minutes; run it in a persistent shell like tmux.

   bash

   1	./scripts/netflow-db.sh pipeline \
   2	--dataset example \
   3	--start-date <YYYY-MM-DD> \
   4	--end-date <YYYY-MM-DD> \
   5	--nfdump target/nfdump/libexec/nfdump

4. Start the dashboard, then open http://localhost:5173.

   bash

   1	bun run dev:web

If a step fails, read Troubleshooting.

Documentation

• Install and use ATLANTIS
• Develop ATLANTIS
• Domain context and invariants

For a new setup, use these procedures:

• Set up the web applications
• Configure the datasets
• Set up the data pipeline

Acknowledgment

Developed by Oliver Boorstein under support by NSF Research Experiences for Undergraduates with the Oregon Networking Research Group.

Advised by Chris Misa and Reza Rejaie.