Abstract
Inevitable tradeoff between read performance and space saving always shows up when applying offline deduplication for primary storage. We propose Mudder, a multi-tiered and dynamic SLA-driven deduplication framework to address such challenge. Based on specific Dedup-SLA configurations, Mudder conducts multi-tiered deduplication process combining Global File-level Deduplication (GFD), Local Chunk-level Deduplication (LCD) and Global Chunk-level Deduplication (GCD). More importantly, Mudder dynamically regulates deduplication processes according to instant workload status and predefined Dedup-SLA during runtime.