作者: Thorsten Schütt , Florian Schintke , Alexander Reinefeld
关键词:
摘要: We present nsync, a tool for synchronizing large replicated data sets in distributed systems. nsync computes nearly optimal synchronization plans based on hierarchy of gossip algorithms that take the network topology into account. Our primary design goals were maximum performance and scalability. achieved these by exploiting parallelism planning phase, omitting transfer unnecessary metadata, at block level rather than file level, using sophisticated compression methods. With its relaxed consistency semantic, neither needs master copy nor quorum updating replicas. Each replica is kept as an autonomous entity can be modified with usual tools.