Shard Manager: A Generic Shard Management Framework for Geo-distributed Applications

ACM Symposium on Operating Systems Principles (SOSP)

Abstract

Sharding is widely used to scale an application. Despite a decade of effort to build generic sharding frameworks that can be reused across different applications, the extent of their success remains unclear. We attempt to answer a fundamental question: what barriers prevent a sharding framework from getting adopted by the majority of sharded applications?

We analyze hundreds of sharded applications at Facebook and identify two major barriers: 1) lack of support for geo-distributed applications, which account for most of Facebook’s applications, and 2) inability to maintain application availability during planned events such as software upgrades, which happen ≈ 1000 times more frequently than unplanned failures. A sharding framework that does not help applications to address these fundamental challenges is not sufficiently attractive for most applications to adopt it. Other adoption barriers include the burden of supporting many complex applications in a one-size-fit-all sharding framework and the difficulty in supporting sophisticated shard-placement requirements. Theoretically, a constraint solver can handle complex placement requirements, but in practice it is not scalable enough to perform near-real-time shard placement at a global scale.

We have overcome these adoption barriers in Facebook’s sharding framework called Shard Manager. Currently, Shard Manager is used by hundreds of applications running on over one million machines, which account for about 54% of all sharded applications at Facebook.

Latest Publications

Sustainable AI: Environmental Implications, Challenges and Opportunities

Carole-Jean Wu, Ramya Raghavendra, Udit Gupta, Bilge Acun, Newsha Ardalani, Kiwan Maeng, Gloria Chang, Fiona Aga Behram, James Huang, Charles Bai, Michael Gschwind, Anurag Gupta, Myle Ott, Anastasia Melnikov, Salvatore Candido, David Brooks, Geeta Chauhan, Benjamin Lee, Hsien-Hsin S. Lee, Bugra Akyildiz, Max Balandat, Joe Spisak, Ravi Jain, Mike Rabbat, Kim Hazelwood

MLSys - 2022