Akamai’s SRE team tackled the thorny task of scheduling disruptive hypervisor maintenance across a fleet that serves hundreds of thousands of VMs. The problem blends capacity limits, migration concurrency, and strict customer‑disruption SLAs, forcing engineers to juggle spare compute, network bandwidth, and conflict constraints. After trialing several commercial and open‑source MIP solvers, they settled on Google’s OR-Tools library.

CP‑SAT, the constraint‑programming engine inside OR‑Tools, proved a natural fit because it natively supports interval variables and cumulative resources. Modeling each VM migration as a ten‑unit interval allowed the team to enforce the AddNoOverlap and AddCumulative constraints that respect the three‑C limits. In effect, the maintenance schedule maps onto a Resource‑Constrained Project Scheduling Problem (RCPSP) without precedence links.

The prototype runs on a single host with three VMs and a 100‑unit planning horizon, yet scales to datacenter‑wide fleets by adding more interval variables and adjusting the cumulative capacity parameter. Engineers can now generate schedules that finish maintenance faster while keeping migration load under control, delivering a measurable reduction in customer‑visible downtime. Akamai plans to integrate the solver into its production orchestration pipeline.