Fast Data Transport and Distribution
A research dataset needs to reach several sites; a data stream has to remain useful across a long network path. FILA investigates how transport protocols, congestion control and distribution structures work together. High bandwidth and long delays make the task challenging: considerable data may already be in flight before feedback reaches the sender.
RMDT brings multiple receivers into one transport session. Application-layer multicast arranges participating applications into a forwarding tree. Two decisions are therefore connected: how fast to send, and through which applications the data reaches its receivers. Sender adaptation, queues and recovery of missing packets are studied together.
Evaluation connects the rate of useful data delivered with delay, loss and resource demand. The setup, background traffic, end systems and protocol settings make results interpretable. LTest exposes individual packet sequences; HighPerTimer supports local timing operations. The bandwidth-delay product puts a scale on the data that can be in flight at once.
AI-RMDT extends these questions towards learned transport control. The 2026 PPO study transferred a controller trained at moderate bandwidth to a real WAN path with a 10 Gbit/s interface budget. In the analysed single-flow run, the mean receive rate was 49.6 Mbit/s and the peak 438 Mbit/s. Oscillation and packet loss showed that a trained policy does not automatically scale to a much larger bandwidth-delay range.
The study examines one path, with timing and implementation limits. CUBIC and BBR checked path capacity using four parallel flows; these are not matched single-flow or fairness comparisons. The findings motivate scale-aware training and more precise packet pacing. Read the PPO scaling study’s setup and results. Contact the FILA team to discuss experiments or student topics.
Selected publications
- Nikolai Mareev, Dmitry Kachan, Eduard Siemens (2022). Multipoint Data Transmission Issues in High Bandwidth-Delay Product TCP/IP Networks. Progress in Advanced Information and Communication Technology and Systems, Lecture Notes in Networks and Systems 548, 277–295 (Springer) DOI: 10.1007/978-3-031-16368-5_14.
- Ali Ghermezian, Kirill Karpov, Dmitry Kachan, Veronika Kirova, Eduard Siemens (2026). PPO-Driven Reinforcement Learning Congestion Control Under High-BDP Wide-Area Deployment: A Scaling Analysis. Proceedings of the International Conference on Applied Innovations in IT (ICAIIT), 14(2), 43–49 DOI: 10.25673/123807.
Research Areas · Team and Contact
Explore theses and student projects at FILA, or discover topics for your own work.
