Detail

Blotter: Scalable and Low-Latency ACID Transactions for Geo-Replicated Storage

Most geo-replicated storage systems are based on weak consistency models for avoiding the performance penalty of coordinating replicas in different data centers. This departure from stronger semantics, such as ACID, poses problems to application programmers, who need to address the issues associated with weak consistency in applications. Recent proposals for transactional semantics in geo-replicated systems exist, but they either require a departure from conventional semantics, or multiple cross-datacenter round-trips.

In this talk we will present Blotter, which, to the best of our knowledge, is the first geo-replicated system to support generic ACID transactions with Snapshot Isolation (SI), only requiring a single cross-datacenter round-trip for coordination at commit time. Blotter combines a novel version-based definition of SI, a new concurrency control protocol that leaves a small amount of local state during reads to make commits more efficient, and a new configuration of Paxos, tailored for good performance in wide area settings. We implemented Blotter as an extension to the widely used Cassandra NoSQL storage system. Our experimental evaluation using multiple datacenters shows the overhead introduced by Blotter to be negligible at a geo-replication scale while ensuring ACID transaction semantics.