Bandana: Using Non-volatile Memory for Storing Deep Learning Models
This addresses the memory bottleneck for recommender systems by reducing DRAM dependency, though it is an incremental improvement in storage optimization.
The paper tackles the high DRAM cost of storing embeddings in large-scale recommender systems by introducing Bandana, a storage system that uses Non-volatile Memory (NVM) as primary storage with DRAM caching, achieving a 2-3x increase in effective NVM read bandwidth and reducing total cost of ownership.
Typical large-scale recommender systems use deep learning models that are stored on a large amount of DRAM. These models often rely on embeddings, which consume most of the required memory. We present Bandana, a storage system that reduces the DRAM footprint of embeddings, by using Non-volatile Memory (NVM) as the primary storage medium, with a small amount of DRAM as cache. The main challenge in storing embeddings on NVM is its limited read bandwidth compared to DRAM. Bandana uses two primary techniques to address this limitation: first, it stores embedding vectors that are likely to be read together in the same physical location, using hypergraph partitioning, and second, it decides the number of embedding vectors to cache in DRAM by simulating dozens of small caches. These techniques allow Bandana to increase the effective read bandwidth of NVM by 2-3x and thereby significantly reduce the total cost of ownership.