Understanding Data Storage and Ingestion for Large-Scale Deep Recommendation Model Training
Authors:
- Mark ZhaoMZMark Zhao
- Niket AgarwalNANiket Agarwal
- Aarti BasantABAarti Basant
- Bugra GedikBGBugra Gedik
- Satadru PanSPSatadru Pan
- Mustafa OzdalMOMustafa Ozdal
- Rakesh KomuravelliRKRakesh Komuravelli
- Jerry PanJPJerry Pan
- Tianshu BaoTBTianshu Bao
- Haowei LuHLHaowei Lu
- Sundaram NarayananSNSundaram Narayanan
- Jack LangmanJLJack Langman
Preprint
Abstract
Datacenter-scale AI training clusters consisting of thousands of domain-specific accelerators (DSA) are used to train increasingly-complex deep learning models. These clusters rely on a data storage and ingestion (DSI) pipeline, responsible for storing exabytes of training data and serving it at tens of terabytes per second. As DSAs continue to push training efficiency and throughput, the DSI pipeline is becoming the dominating factor that constrains the overall training performance and capacity. Innovations that improve the efficiency and performance of DSI systems and hardware are urgent, demanding a deep understanding of DSI characteristics and infrastructure at scale. This paper presents Meta's end-to-end DSI pipeline, composed of a central data warehouse built on distributed storage and a Data PreProcessing Service that scales to eliminate data stalls. We characterize how hundreds of models are collaboratively trained across geo-distributed datacenters via diverse and continuous training jobs. These training jobs read and heavily filter massive and evolving datasets, resulting in popular features and samples used across training jobs. We measure the intense network, memory, and compute resources required by each training job to preprocess samples during training. Finally, we synthesize key takeaways based on our production infrastructure characterization. These include identifying hardware bottlenecks, discussing opportunities for heterogeneous DSI hardware, motivating research in datacenter scheduling and benchmark datasets, and assimilating lessons learned in optimizing DSI infrastructure.Keywords
All Related Versions
This publication has 35 references indexed in Scilit:
- Graphcore2021
- SambaNova SN10 RDU:Accelerating Software 2.0 with Dataflow2021
- Understanding Training Efficiency of Deep Learning Recommendation Models at Scale2021
- The Architectural Implications of Facebook's DNN-Based Personalized Recommendation2020
- Accelerating Data Loading in Deep Neural Network Training2019
- Efficient User-Level Storage Disaggregation for Deep Learning2019
- AutoAugment: Learning Augmentation Strategies From Data2019
- Applied Machine Learning at Facebook: A Datacenter Infrastructure Perspective2018
- Microsoft COCO: Common Objects in Context2014
- ImageNet: A large-scale hierarchical image database2009