Consistent and durable data structures for non-volatile byte-addressable memory

摘要: The predicted shift to non-volatile, byte-addressable memory (e.g., Phase Change Memory and Memristor), the growth of "big data", subsequent emergence frameworks such as memcached NoSQL systems require us rethink design data stores. To derive maximum performance from these new technologies, this paper proposes use single-level For systems, where no distinction is made between a volatile persistent copy data, we present Consistent Durable Data Structures (CDDSs) that, on current hardware, allows programmers safely exploit low-latency non-volatile aspects technologies. CDDSs versioning allow atomic updates without requiring logging. same scheme also enables rollback for failure recovery. When compared memory-backed Berkeley DB B-Tree, our prototype-based results show that CDDS B-Tree can increase put get throughput by 74% 138%. Cassandra, two-level store, Tembo, enabled distributed Key-Value system, increases up 250%-286%.