DVS: Distributed Volatile Storage for low-latency read access

We describe the design and implementation of DVS, a distributed, fault-tolerant in-memory storage system for fast random access to large datasets. The storage has a key-value interface and is optimized for read performance. It is used as a cache for static data stored on slower durable mediums to accelerate access to large datasets, such as those used in bioinformatics. The key problems that DVS solves are data location on and data distribution to a dynamic set of servers. We show that DVS can speed up a real bioinfor- matics application by more than 100%.