Solving virtual machine puzzles: How AI is optimizing cloud computing

Google Research · 9 months ago

Google developed machine learning algorithms called NILAS, LAVA, and LARS to improve how virtual machines are allocated across physical servers in data centers by predicting VM lifetimes and continuously updating those predictions as VMs run. The NILAS algorithm has been deployed in Google's production data centers since early 2024 and increased empty hosts by 2.3–9.2 percentage points while reducing CPU stranding by approximately 3%. These improvements allow data centers to use their computing resources more efficiently and free up capacity for system maintenance and large workloads.