Cache Augmented Data Stores: Solutions and Opportunities
First floor of Chemistry department, Jaber ibn Hayyan Hall
Wednesday, 27 December 2017
09:30 - 10:30
Numerous organizations augment a data store with a cache to enhance application performance. Redis and Memcached are two popular cache managers in use by popular internet destinations including Facebook, YouTube, and Wikipedia. One may purchase these cache managers as a service from a cloud provider such as Amazon AWS, Microsoft Azure, and Google Cloud. In this talk, I provide an overview of how these caches are used. I present today's solutions and future research opportunities, outlining the research activities we are pursuing at the USC database laboratory. Solutions include CAMP as a smart cache space management technique, alternative write policies including their tradeoffs, the Inhibit, and Quarantine leases to provide strong consistency, support for range predicates using RangeQC, data migration, and programmable switches to resolve hotspots and bottlenecks. Opportunities include support for the emerging Non-Volatile Memory (NVM), auto-scaling of cache size based on demand, and frameworks that make these caches transparent. I present the system methodology we use to build prototypes and how we establish correctness using frameworks such as Polygraph.
Shahram Ghandeharizadeh received his Ph.D. degree in Computer Science from the University of Wisconsin, Madison, in 1990. Since then, he has been on the faculty at the University of Southern California where he is the director of the Database Laboratory. He has received numerous awards and recognitions for his research and teaching including the National Science Foundation Young Investigator’s Award, USC School of Engineering Research and Outstanding Teaching Awards,and the ACM Software System Award. He has published more than a hundred articles and organized numerous conferences. He is the co-inventor of the BG benchmark and the founder of Mitra, a startup commercializing a high performance cache-augmented data store named KOSAR.