Research On Smart Cloud Disk Allocation Policy In Cloud Storage System

Cloud Block Storage(CBS)system is a mainstream storage architecture in the current cloud storage system.Cloud providers provide virtual block-level storage de vices(called cloud disks)for their tenants by setting up cloud block storage syste ms.In recent years,the types of workloads and the amount of data stored on the cloud have increased rapidly,with the number of cloud disks reaching millions,lea ding to increased utilization of cloud block storage system storage clusters,which brings huge challenges for the resource management of cloud block storage system.One of the main ways of resource management for cloud block storage systems is to allocate cloud disks to appropriate storage clusters.Due to the lack of in-depth analysis of user workloads,traditional allocation policies only consider a single resource dimension or a single optimization goal,which leads to load imbalance,low average utilization of storage resources,and high data migration traffic.To this end,the smart placement policy Smart-P and the smart migration policy Smart-M are proposed.For Smart-P,to solve the problem of lack of user load information during placement,a load prediction algorithm based on the decision tree model is proposed to predict future workload when a new cloud disk is created,and then,based on the predicted load information,cloud disk placement is performed using a placement policy that minimizes resource utilization conflicts.Compared with the existing one-dimensional allocation policy,Smart-P can improve the resource utilization of the cluster by about 30% and reduce the amount of migrated data by60%.In Smart-M,an improved cloud disk migration decision-making algorithm based on a multi-objective evolutionary algorithm is proposed,which integrates multiple objectives such as load balancing,peak traffic dispersion,and data migration traffic reduction,solving the problem that traditional methods cannot effectively optimize.Smart-M uses a greedy heuristic algorithm to select cloud disks which could be migrated and performs population initialization by simulated annealing algorithm to reduce the time overhead of the decision process.Compared with the traditional heuristic policy,the Smart-M migration policy resulted in a 55% reduction in cluster overload time,90% reduction in cluster overload degree,and 30% reduction in data migration amount.