Scientific Computing Seminar

In this talk, I will present Summary Hash History (SHH), a novel decentralized ordering mechanism that provides secure and scalable optimistic replication. Being based on a causal history approach with secure summary hashes as version identifiers, SHH supports simple management of site membership changes, scales regardless of the number of sites, and guarantees the correctness of decentralized ordering in a scalable way. SHH uses "two-step reconciliation" to overcome the inherent limitation of the causal history approach, and thus, consumes orders of magnitude lower bandwidth than reconciliation based on version vectors. Interestingly, perhaps surprisingly, SHH provides faster convergence than version vector based approaches by recognizing "coincidental equalities," cases when identical versions are produced independently at different sites. This is of significant value in that SHH can enable distributed replica to converge even in the partitioned network.