EAGER: Elastic Multi-layer Memcached Tiers

Facebook and YouTube have massive databases containing many objects. Objects (e.g., pictures, videos) are read from these databases and presented to users. These databases use slower and cheaper I/O devices. Repeated database queries are also slow, so companies cache popular objects in volatile memory (RAM). As RAM is about 1000x faster than I/O, queries to popular objects are serviced faster. A single computer cannot efficiently use too much RAM, so a distributed memory caching system called "Memcached" is utilized. Memcached creates a cluster of nodes, forming a key-value database in faster memory. Memcached's simple architecture led to its popularity, but is also its Achilles' heel. When loads increase, more caching nodes are needed, but it is difficult to add new nodes and redistribute the cached data to more nodes. Similarly, when loads reduce, it is harder to shut down some nodes and migrate their cached data to fewer nodes; scaling down the cluster helps reduce the large energy costs they consume. Any changes to the number of Memcached nodes result in throwing out most cached data, leading to lengthy periods where the distributed caches have to be slowly re-warmed up from the backend database. This EArly-concept Grants for Exploratory Research (EAGER) project investigates techniques to scale Memcached clusters smoothly, without any transient performance degradations. The project also introduces an intermediate Flash-based storage tier between the memory nodes and the backend database, to help reduce high latencies to the backend database and help the smooth scaling of the Memcached cluster.