GoSeed: Generating an Optimal Seeding Plan for Deduplicated Storage
summary
GoSeed formulates seeding as an ILP formulation and ships three practical accelerations: solver timeout, fingerprint sampling, and container aggregation. On three real workloads GoSeed consistently finds feasible plans when greedy baselines sometimes fail, cuts replication cost dramatically on high-duplication cases (200 daily backups is the hardest), and often yields the lowest replication fraction. Sampling speeds solver time by orders of magnitude so full ILP solves take minutes rather than hours while container aggregation avoids unpacking and reduces problem size.
AI-written summary · about