VectorCDC: Accelerating Data Deduplication with Vector Instructions

This conference talk from USENIX FAST '25 presents VectorCDC, a novel method for accelerating Content-defined Chunking (CDC) algorithms using SSE/AVX CPU vector instructions. Learn how researchers from the University of Waterloo address a major performance bottleneck in data deduplication systems by implementing hashless CDC with vector acceleration. The 12-minute presentation demonstrates how VectorCDC achieves 21-46× higher throughput compared to existing vector acceleration techniques while maintaining the same space savings efficiency. Particularly valuable for storage system engineers, computer scientists, and professionals interested in data deduplication optimization techniques.