QBox: guaranteeing I/O performance on black box storage systems
Proceedings of the 21st international symposium on High-Performance Parallel and Distributed Computing
| Hi-index | 0.00 |
Guaranteed I/O performance is needed for a variety of applications ranging from real-time data collection to desktop multimedia to large-scale scientific simulations. Reservations on throughput, the standard measure of disk performance, fail to effectively manage disk performance due to the orders of magnitude difference between best-, average-, and worst-case response times, allowing reservation of only a small fraction of the achievable bandwidth. Moreover, hard I/O performance guarantees for a mix of workloads are generally considered impractical due to the stateful nature of disk I/O and the interference between workloads. We present a solution for hard I/O performance guarantees for a mix of workloads with different performance and timeliness requirements without sacrificing performance. This is achieved by reserving and managing disk resources in terms of disk time utilization. Our novel disk I/O scheduler, Fahrrad, makes hard guarantees on disk time utilization reservation while maintaining high I/O performance. We characterize and quantify all overheads introduced by seeks between competing workloads, and introduce a new method for quantifying, reserving, and accounting for these overheads to provide complete isolation between workloads. The resulting system provides "virtual disks" to clients, providing the illusion of solo access to a disk with exactly the desired performance. Moreover, by making a simple translation from workload performance requirements and behaviors into the disk time utilization, our system provides arbitrary hard or soft throughput and latency guarantees depending upon application needs. Making hard performance guarantees for multiple disks require assumptions about the layout of data to make disk reservations. This often leads to poor utilization of disk resources, making it challenging to effectively manage performance of multiple disks We present a decentralized, multi-layered solution that locally manages individual disk resources and delivers soft throughput and latency targets without relying on disk reservations. By applying our principles of managing disk I/O in terms of disk time utilization, we develop a novel disk I/O scheduler, Horizon, which meets individual request deadlines while achieving high disk performance.