Sedna: A Memory Based Key-Value Storage System for Realtime Processing in Cloud

摘要

Comparing with the traditional disk based distributed storage system, RAM based storage has been proven to be an effective way to accelerate real time applications processing speed. In this paper, we propose a memory based distributed Cloud storage system called Sedna. Managing 'big data' across lots of commodity servers, Sedna provides high scalability, simple effective data access APIs with data consistency and persistency, and a new trigger based APIs for real time applications. To guarantee the scalability with low latency, we design and implement a hierarchical structure to manage huge size data center which is simple and effective. Except the high speed provided by memory based storage, Sedna absorbs advantages from state-of-art cloud programming frameworks, and gives programmers a new way to write massive data real time applications. These data read/write triggers APIs are necessary but are missing parts of modern distributed storage system. Experiments and examples show Sedna achieve comparable speed to widely-used distributed cache system, and provide a more efficient way to use distributed storage.