BSc Research Project: Autotuning Geo-Distributed Systems

In this project I supervise 5 BSc students in implementing and evaluating different autotuning policies for geo-distributed systems. Each student focuses on one autotuning mechanism.

Each student implements and tests one of the autotuning policies following this research method:

1. Study existing autotuning systems’ strategies
2. Profile the performance and identify bottlenecks of Detock and Styx
3. (For reinforcement learning policies) Collect training data from manual rescaling/data movement decisions
4. Design and implement a new policy
5. Test and evaluate performance improvement (throughput & latency) and cost savings

Policies to Implement

• A control-based policy for automatic resource allocation
• A time series forecasting policy for automatic resource allocation
• A control-based policy for dynamic data movement
• A reinforcement learning-based policy for dynamic data movement
• A privacy-preserving policy fot dynamic data movement

Students

1. Rares Popa - TU Delft
2. Mihai Nicolae - TU Delft
3. Arpad Jakab - TU Delft
4. Kevin Che - TU Delft
5. Frank Verkoren - TU Delft

Final reports coming soon