Accelerating duplicate data chunk recognition using NN trained by locality-sensitive hash

Deduplication is often used in storage systems in order to save storage space, communication bandwidth, write energy, and recovery and error-protection infrastructure. However, deduplication overhead increases latency and computation energy. Determining whether a data chunk is already stored by comparing signatures constitutes a significant fraction of this deduplication overhead. In this paper, we propose a statistical chunk classifier based on a neural network. Our technique is based on learning the patterns of locality-sensitive hashing of the data. Our experiments show an acceleration of chunk processing, leading to reduction in deduplication overhead.