期刊名称：Bulletin of the Technical Committee on Data Engineering
出版年度：2017
卷号：40
期号：1
页码：62
出版社：IEEE Computer Society
摘要：RAMCloud is a scale-out data center key-value store designed to aggregate the DRAM of thousandsof machines into petabytes of storage with low 5 s access times. RAMCloud was an early system inthe space of low-latency RDMA-based storage systems. Today, it stands as one of the most completeexamples of a scalable low-latency store; it has features like distributed recovery, low fragmentationthrough a specialized parallel garbage collector, migration, secondary indexes, and transactions.This article examines RAMCloud’s key networking and dispatch decisions, and it explains how theydiffer from other systems. For example, unlike many systems, RAMCloud does not use one-sided RDMAoperations, and it does not partition data across cores to eliminate dispatch and locking overheads.We explain how major RAMCloud functionality still exploits modern hardware while retaining aloose coupling both between a server’s internal modules and between servers and clients by explainingour recent work on Rocksteady, a live data migration system for RAMCloud. Rocksteady uses scat-ter/gather DMA and multicore parallelism without changes to RAMCloud’s RPC system and with mini-mal disruption to normal request processing.
