โ† Back to the chapter

Technical detail ยท Fog Resource Allocation

A stable allocation policy

17.6 The Stable Allocation Model

A stable allocator is easier to audit when hard gates are separated from ranking. One practical model is:

1. Protect reserves:
   keep control_reserve, recovery_reserve, and observability_reserve available

2. Reject infeasible work:
   deadline_ok && memory_ok && network_ok && storage_ok &&
   energy_ok && thermal_ok && fallback_ok

3. Admit work by class:
   critical: reserve and priority queue
   important: quota plus deadline monitoring
   best_effort: spare capacity only

4. React to pressure:
   throttle best_effort before important work
   shed optional work before violating critical deadlines
   migrate only when destination headroom and state cost are acceptable

5. Review the decision:
   reopen when deadline misses, queue growth, thermal throttling,
   flash pressure, retry storms, or repeated failovers appear

The model can use weighted fairness, deadline scheduling, token buckets, priority queues, or container limits. The important part is that the chosen mechanism is tied to the workload contract and verified under pressure.

No high score overrides a hard gate. Spare CPU cannot compensate for insufficient memory headroom, control reserve, storage endurance, thermal budget, or fallback behavior.

โ† Back to the chapter