报告题目：ElasticOnline Analytical Processing on RAMCloud

报 告 人：刘秋彤

报告时间：2014年4月25日

报告地点：贵州大学北校区博学楼601室

报告内容摘要：

A shared-nothing architecture is state-of-the-artfor deploying a distributed analytical in-memory database management system: itpreserves the in-memory performance advantage by processing data locally oneach node but is difficult to scale out. Modern switched fabric communicationlinks such as InfiniBand narrow the performance gap between local and remoteDRAM data access to a single order of magnitude. Based on these premises, weintroduce a distributed in-memory database architecture that separates thequery execution engine and data access: this enables a) the usage of a large-scaleDRAM-based storage system such as Stanford’s RAMCloud and b) the push-down ofbandwidth-intensive database operators into the storage system. We address theresulting challenges such as finding the optimal operator execution strategyand partitioning scheme. We demonstrate that such an architecture delivers both:the elasticity of a shared-storage approach and the performance characteristicsof operating on local DRAM.