| Currently,the demand for edge-side data and proximity access to data services is exploding,and the big data processing model of cloud computing can no longer meet the latency demand of users for data processing.Therefore,the edge computing paradim,which is closer to the user side,is expected to meet this demand.On the other hand,there is a growing trend to use scattered,limited,and heterogeneous edge-side resources as an overall resource pool to provide efficient services.With the influx of massive edge data into an overall heterogeneous edge storage system,there is a challenge to allocate and store the data among these heterogeneous edge nodes in order to fully utilize the node storage resources and evenly distribute the system storage load.In addition,the storage nodes where some of the hot data are located will be frequently requested,making the access load of these nodes too high.Load balancing capability is the key to a system’s stability and continuity of service.In order to ensure that the heterogeneous edge storage system can provide data storage and access services to users stably and continuously,this paper conducts two researches on storage and access load balancing.(1)A storage load distribution method is designed to solve the storage load balancing problem of the heterogeneous edge storage system.The method allocates storage load from a global view through a software-defined network.Specifically,the heterogeneity of each storage node is measured at the control level,then the Delauany and weighted Voronoi graph structures are constructed to achieve storage load balancing among heterogeneous edge nodes,and finally the corresponding forwarding policies are formulated for this virtual structure.In this way,the massive data entering the system can be evenly distributed according to the capacity of edge nodes,and the overall efficiency of storage resource utilization can be improved.The simulation results based on Mininet show that this method has better resource utilization and storage load balancing effect than two existing representative schemes such as GRED and Chord.(2)The distributed edge storage system faces a very uneven distribution of data query requests,and a large number of data queries will be concentrated on a small number of edge nodes storing hot data.This results in a high access load on the edge nodes storing these data and a high load on the network links and devices through which the query requests pass,while the access load on other edge nodes is relatively low.The unbalanced access load of the system will cause the system operation efficiency and stability to be greatly reduced,and the actual user experience will be poor.In this paper,we propose an access load balancing strategy based on the prediction of the distribution of query requests,which predicts the hot data,creates replicas for the hot data at the appropriate locations,and finally sets access rules and update policies for the newly added replicas.Through these two methods,the load balancing problem of heterogeneous edge storage system can be effectively solved.The utilization of node resources in the system can be maintained at a high level without causing idle storage resources,and access requests can be effectively distributed to other nodes in the system.This paper focuses on the load balancing problem of heterogeneous edge storage systems and does not address some other problems in network transmission.In addition,the load balancing decision relies on centralized processing at the control level,but centralized processing brings additional time overhead,hence these issues will be weighed together with load balancing optimization in the subsequent practical system improvement. |
