Research On Quality Of Service Guarantee Of Storage System Based On Dynamic Scheduling Strategy

High performance computing systems are beginning to enter the exascale era.More and more data-intensive applications run on a large-scale in high-performance computing systems,resulting in an increasing demand for capable storage systems.At the same time,HPC is migrating to the cloud,and multi-tenant and multi-load scenarios are more common in Cloud HPC systems,which means that storage resources in HPC system are shared between more and more different users and limited storage resources need to serve more applications.When multiple applications access the storage service concurrently,they will compete for I/O resources and cause I/O interference,which will lead to a serious drop in I/O aggregate bandwidth,the extension of applications runtime and severe performance degradation.In the meanwhile,hybrid storage architecture is also widely used in HPC environment.Therefore,it is vital that how to use the characteristics of storage devices to ensure the quality of service of applications with limited physical resources.In order to solve these challenges,we propose DDL-QoS,a quality of service guarantee method based on dynamic scheduling strategy.DDL-QoS do not blindly pursue maximum performance,and guarantee quality of service in forms of meeting requirements of applications which is similar to the on-demand service in cloud computing.We use deadline to represent the requirements of applications and deadline refers to the runtime of an application running separately without interference.Firstly,the quality of service guarantee method based on dynamic scheduling strategy explore the bandwidth allocation phenomenon caused by interference in applications through the analyzing of historical data,and then proposes a calculation formula according to this regularity distribution,which can be used to dynamically perceive the load capacity of storage devices.Secondly,quote a metric called random percentage that can represent the random degree of the applications and be used to guide I/O scheduling in the later stage.At last,DDL-QoS uses the above two parameters to synthetically consider the application characteristics and the load capacity of storage devices,and chooses appropriate storage devices to schedule I/O,so as to reduce interference,ensure the applications to meet their deadline,and improve the overall performance.Through extensive experiments and analysis,we validate that DDL-QoS can improve the performance by an average of 3.02 times under our experimental environment.