Over the past several decades, DRAM and hard disks have been the dominating computer memory and storage technologies. In recent years, emerging non-volatile memory (NVM) technologies such as flash are orders of magnitude faster than hard disks. Next-generation NVMs such as phase change memory and the memristor are byte-addressable and promise to have close-to-DRAM performance. These NVM technologies are poised to radically alter the performance landscape for storage systems, blurring the line between memory and storage. To fully utilize them and deliver their good performance to applications, system designers need to rethink how to efficiently manage them as reliable storage at different layers.
In this talk, I will discuss new challenges NVM technologies present to system designers and the systems that I have built to efficiently manage them. My approach is to vertically rethink different layers in a system and optimize them in a coherent way to deliver all the capabilities of the new NVM technologies. Specifically, I will present Mojim, a system that provides reliable and highly-available non-volatile main memory in data center environments. Mojim uses a flexible architecture and efficient software and networking stacks to achieve good data replication performance that is close to or even better than un-replicated single-node system. I will also present another system that I built for flash-based solid state drives (SSDs), where I remove excess indirection and its overhead in current SSDs. Finally, I will discuss my future research plans in building systems for “little” big data.
Speaker Biography
Yiying Zhang is a postdoctoral scholar in the Department of Computer Science and Engineering at the University of California, San Diego, where she works with Professor Steven Swanson in the Non-Volatile Systems Lab. Her research interests span operating systems, distributed systems, computer architecture, networking, and data analytics, with a focus on building fast, reliable, and flexible systems for emerging hardware and applications. She is currently interested in system implications of non-volatile main memory. In the past, she has worked in various aspects of storage systems, including removing excess indirection in storage systems, storage-level caching, and redundant arrays. Yiying received her Ph.D. from the Department of Computer Sciences at the University of Wisconsin-Madison in 2013, under the supervision of Professors Remzi Arpaci-Dusseau and Andrea Arpaci-Dusseau.