As rapid advancement of storage R&D, storage technologies have become increasingly diverse (e.g., hard drives, SSDs, and non-volatile memory), which allows storage systems to provide different qualities of storage services (QoSS). In the meantime, data-intensive and big data applications, such as database and cloud storage systems, show increasingly high demand for differentiated storage services, in terms of performance, reliability, power efficiency and others. Unfortunately, the existing I/O infrastructure and OS design are still centered on the decade-old block-based interface, which does not allow communicating the QoSS needs between service consumers (applications) and service providers (storage systems). Such a disconnection results in (1) sub-optimal data management in storage systems, due to the lack of semantic information from applications, and (2) a huge waste of optimization opportunities for applications, due to the over-simplified storage model. In this talk, I will share our thoughts on this fundamental issue and present our system solution. We have designed and implemented a complete stack, including an optimized PostgreSQL database system, an instrumented Linux OS kernel, and a tiered storage system with SSD caching. Our research shows that with only slight system changes, enabling application-awareness can significantly improve application performance and optimize data management in storage systems, which can timely and effectively respond to the increasingly high storage service demand of data-intensive applications.
Feng Chen is a research scientist with Intel Labs, Oregon. His current research interests focus on computer systems areas, including operating systems, storage systems, and big data management in cloud and large-scale distributed systems. Besides publications in major venues, several of his research work has been adopted by open-source communities and influenced industrial products, such as the CLOCK-Pro algorithm (USENIX’05) and the Hystor storage system (ICS’11). In 2010, he received Mike Liu CSE Graduate Research Award at Ohio State. He is also a recipient of the Best Paper Award in the 25th ACM International Conference on Supercomputing in 2011.
Feng Chen received his Ph.D. degree (2010) and M.S. degree (2009) in Computer Science and Engineering at the Ohio State University. He received his M.E. degree (2003) and B.E. degree (2000) in Computer Science at Zhejiang University, China.