Linearizable Low-latency Reads at the Edge

Abstract

Edge computing enables moving data closer to users to reduce latency and improve user experience. Edge data centers are capable and reliable enough to support various data management solutions, such as caches and data stores. Unfortunately, edge storage systems sacrifice consistency to benefit from geographical proximity to users. In this paper, we present EdgePQR, a strongly consistent, edge-aware data store that allows edge clients to read "hot" data locally at the edge with low latency. EdgePQR relies on the piece-wise defined quorums consisting of nodes in a core cloud system and one or more edge data centers to replicate data to the edge. It then uses an edge quorum to query data locally. EdgePQR ensures safety by enforcing intersections between all vital quorums: any leader election and replication quorums intersect, and any replication and edge-read quorums intersect as well.