CARA: Exact Local Repair Queries for Cloud Consolidation
Cloud consolidation evaluates alternative workload placements to reduce overload under fixed host and migration budgets. A controller repeatedly queries the same demand history, but consumes only a small ranked result and one final repair. Shared aggregation and incremental assignment reuse individual computations but still reconstruct constraints and numerical certificates for many related queries. This preparation can erase the savings from skipping exact profiles. We present CARA, which compiles seed-universal movement constraints and family-wide load-error budgets, then acquires exact profiles according to certified ranking keys. The compiler leaves a migration-bounded assignment core and preserves ranked records, physical mappings and continuation state; a separate admission stage evaluates one frozen repair. On all 12,006 hourly series from four workload sources, CARA executes controlled queries 4.75 times faster than compact complete profiling and heterogeneous packed queries 3.05 times faster than family-local execution; across 300 budgeted consolidation decisions, it reduces excess over FFD by 35.8% and attains the lowest excess among the compared policies.