Dimitris Skourtis, Noah Watkins, Dimitris Achlioptas, Carlos Maltzahn, Scott Brandt
07/18/2013 09:56 AM
Computer Science
Modern applications and virtualization require fast and predictable storage. Hard-drives have low and unpredictable performance, while keeping everything in DRAM, in many cases, is still prohibitively expensive or unnecessary. Solid-state drives offer a balance between performance and cost, and are becoming increasingly popular in storage systems, playing the role of large caches and permanent storage. Although their read performance is high and predictable, under read/write workloads solid-state drives frequently block and exceed hard-drive latency.
In this paper, we propose an efficient approach for achieving performance predictability in distributed storage systems comprised of solid-state drives. By observing that virtually all storage systems incorporate significant redundancy for the purpose of reliability, we propose exploiting this latent resource to achieve the separation of reads from writes across nodes, allowing each drive to periodically serve either read or write workloads. Our proposed approach provides high performance and low latency for reads under read/write workloads while adding no extra cost to the system.
