Eris: Coordination-Free Consistent Transactions Using In-Network Concurrency Control

摘要: Distributed storage systems aim to provide strong consistency and isolation guarantees on an architecture that is partitioned across multiple shards for scalability replicated fault tolerance. Traditionally, achieving all of these goals has required expensive combination atomic commitment replication protocols -- introducing extensive coordination overhead. Our system, Eris, takes a different approach. It moves core piece concurrency control functionality, which we term multi-sequencing, into the datacenter network itself. This primitive responsibility consistently ordering transactions, new lightweight transaction protocol ensures atomicity. The end result Eris avoids both overhead: show it can process large class distributed transactions in single round-trip from client system without any explicit between or replicas normal case. provides atomicity, consistency, tolerance with less than 10% overhead throughput 3.6-35x higher latency 72-80% lower conventional design standard benchmarks.