2014/03/10 by Sudip Roy, Łucja Kot, Roy, Sudip +13
Computer Science · #Advanced Data Storage Technologies #Cloud Computing and Resource Management #Databases (cs.DB) #Distributed and Parallel Computing Systems #Distributed systems and fault tolerance #FOS: Computer and information sciences
paper · pdf · doi:10.48550/arxiv.1403.2307
openalex publication_date 2014/03/10 · openalex created_date 2022/10/03 · openalex updated_date 2026/07/28
Datastores today rely on distribution and replication to achieve improved\nperformance and fault-tolerance. But correctness of many applications depends\non strong consistency properties - something that can impose substantial\noverheads, since it requires coordinating the behavior of multiple nodes. This\npaper describes a new approach to achieving strong consistency in distributed\nsystems while minimizing communication between nodes. The key insight is to\nallow the state of the system to be inconsistent during execution, as long as\nthis inconsistency is bounded and does not affect transaction correctness. In\ncontrast to previous work, our approach uses program analysis to extract\nsemantic information about permissible levels of inconsistency and is fully\nautomated. We then employ a novel homeostasis protocol to allow sites to\noperate independently, without communicating, as long as any inconsistency is\ngoverned by appropriate treaties between the nodes. We discuss mechanisms for\noptimizing treaties based on workload characteristics to minimize\ncommunication, as well as a prototype implementation and experiments that\ndemonstrate the benefits of our approach on common transactional benchmarks.\n