Meta's Rebalancer 1.0.4 library provides tools for task placement and optimization in computing environments. The software allows developers to define constraints such as CPU and memory capacity, as well as rack-level placement rules for specific jobs.
In testing, the library successfully resolved a two-task model using both local search and HiGHS solvers, with execution times of 0.8 ms and 0.9 ms respectively. The build explicitly rejects unsupported configurations, such as the default XPRESS solver or Gurobi packages.
A demonstration model involving four 10 GB tasks on two 20 GB hosts showed the solver moving one task in 0.3 ms to resolve capacity issues. The system achieved an optimal state, reducing the broken task count from 1 to 0.
For larger datacenter simulations, the library manages eight hosts organized into racks. It balances CPU utilization while enforcing rules like limiting each job to one task per rack. Tests across three builds confirmed consistent objective outcomes and task movement counts.
