IO request scheduling method and system for reducing access delay in parallel file system

Abstract

The invention discloses an IO request scheduling method and system for reducing access delay in a parallel file system. The method of the invention includes calculating and marking the delay sensitivity of the IO request of a client of a parallel file system accessing a server storage device; Under the condition of competing IO requests, IO requests with higher delay sensitivity are preferentially scheduled to reduce their queuing time in the case of competition congestion and achieve the purpose of reducing delay. The object of the present invention is to reduce the waiting time Tw of some IO requests under the condition of IO request competition, thereby reducing the IO response delay perceived by the client application, which can be used to improve various types of parallel files including large-scale high-performance computer systems. The combined processing performance of the system's computer system for typical workloads with mixed demands of high bandwidth and low IO latency.

Description

technical field [0001] The invention relates to the field of computer operating systems, in particular to an IO request scheduling method and system for reducing access delay in a parallel file system. Background technique [0002] Parallel file system is an important part of high-performance computer. It manages a large number of storage devices and is responsible for scheduling a large number of concurrent IO requests, sending them to storage devices in an orderly manner, and the requests are executed in sequence. Different execution orders of IO requests will reflect different Although there is usually a request queue inside the storage device, the IO queue depth of the storage device is limited. Under large-scale concurrent conditions, the IO request will inevitably be queued at the software layer first. Therefore, the IO request scheduler of the parallel file system layer It has a great impact on the IO response time of the user program and the throughput rate of the st...

AI Technical Summary

Problems solved by technology

In this new load mode, delay-sensitive IO requests and bandwidth-sensitive IO requests are queued in the queue of the parallel file system. When the number of requests is large, competition will occur, which will inevitably lead to a longer waiting time for IO requests to be scheduled. , which impacts the performance of latency-sensitive applications

Under the condition of using traditional disk media, because the disk itself has a large access delay, the proportion of IO request queuing time is relatively low. This problem is not prominent, but when new storage media (for example, new non-volatile storage Volatile Memory) has been widely used, because its access latency is several orders of magnitude lower than that of disks. At this time, in the entire IO access process, the proportion of IO request queuing time will increase sharply, and the impact on application performance will be prominent.

Images