Residential GPU networks turn household hardware and rooftop solar into a distributed computing resource. Making that trustworthy requires more than scheduling: workloads must be isolated, telemetry must be authentic, and load-balancing decisions must not be manipulable by a malicious participant.
Within this project Wei Song leads the security theme, covering:
- secure communication and integrity protection for energy and workload telemetry;
- trustworthy, energy-aware load balancing that resists manipulation and protects participant privacy;
- interfaces between the GPU network and battery and energy management systems studied in the federated BMS project.
The work is connected to the spinout SunstackAI, co-founded by Wei Song, which develops energy-aware load balancing for distributed GPU networks.