Popular SSD-based key-value stores consume a large amount of DRAM in order to provide high-performance database operations. However, DRAM can be expensive for data center providers, especially given recent global supply shortages that have resulted in increasing DRAM costs. In this work, we design a key-value store, MyNVM, which leverages an NVM block device to reduce DRAM usage, and to reduce the total cost of ownership, while providing comparable latency and queries-per-second (QPS) as MyRocks on a server with a much larger amount of DRAM. Replacing DRAM with NVM introduces several challenges. In particular, NVM has limited read bandwidth, and it wears out quickly under a high write bandwidth. We design novel solutions to these challenges, including using small block sizes with a partitioned index, aligning blocks post-compression to reduce read bandwidth, utilizing dictionary compression, implementing an admission control policy for which objects get cached in NVM to control its durability, as well as replacing interrupts with a hybrid polling mechanism. We implemented MyNVM and measured its performance in Facebook's production environment. Our implementation reduces the size of the DRAM cache from 96 GB to 16 GB, and incurs a negligible impact on latency and queries-per-second compared to MyRocks. Finally, to the best of our knowledge, this is the first study on the usage of NVM devices in a commercial data center environment.

Deep learning recommendation models (DLRMs) have been used across many business-critical services at Meta and are the single largest AI application in terms of infrastructure demand in its data-centers. In this paper, we present Neo, a software-hardware co-designed system for high-performance distributed training of large-scale DLRMs. Neo employs a novel 4D parallelism strategy that combines table-wise, row-wise, column-wise, and data parallelism for training massive embedding operators in DLRMs. In addition, Neo enables extremely high-performance and memoryefficient embedding computations using a variety of critical systems optimizations, including hybrid kernel fusion, software-managed caching, and quality-preserving compression. Finally, Neo is paired with ZionEX , a new hardware platform co-designed with Neo's 4D parallelism for optimizing communications for large-scale DLRM training. Our evaluation on 128 GPUs using 16 ZionEX nodes shows that Neo outperforms existing systems by up to 40× for training 12-trillion-parameter DLRM models deployed in production.

No abstract

No abstract