Bandana: Using Non-volatile Memory for Storing Deep Learning Models

TL;DR

Bandana is a storage system that leverages Non-volatile Memory to store deep learning embeddings, reducing DRAM usage and cost while maintaining high read bandwidth through hypergraph partitioning and cache simulation.

Contribution

It introduces a novel NVM-based storage system for embeddings, employing hypergraph partitioning and cache simulation to improve read bandwidth and reduce costs.

Findings

Increases effective read bandwidth of NVM by 2-3x

Reduces DRAM footprint of embeddings significantly

Lowers total cost of ownership for large-scale recommender systems

Abstract

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.