Research On Dynamic Replica Management Strategy In IPFS File System

Besides electronic trading systems such as Bitcoin and Ethereum,decentralized technologies represented by the blockchain are also applied in Web area.The Interplanetary File System(IPFS),for the first time,combines blockchain technology with the Web to implement a decentralized Web framework.Its goal is to replace HTTP and become the basic protocol of the next generation Internet.However,most of the current storage systems use redundant technologies to achieve file availability like replica.IPFS itself does not have data redundancy measures.Instead,it implements a static replica policy through the cluster management tool.However,static policies have problems such as inability to adjust the number of replicas in consistent with user access heat,and imbalanced storage space.Considering below problems,this paper mainly focuses on:(1)The visit heat of files is a time-series dataset.By predicting the visit heat over a period,the system can adjust the number of replicas in advance to cope with the surge in traffic in a short period of time.This paper proposes a file access heat prediction algorithm based on support vector regression(SVR),which can predict the access heat and adjust the replica number based on the change of heat to cope with the sudden increase in user traffic.(2)By studying the file access method of IPFS,an IPFS-based dynamic replica management(IDRM)system is proposed.System is divided into controlling of numbers and credit-based placing policy.First module can periodically adjust the number of replicas according to the file access heat.After the adjustment of the number of replicas,access latency in high concurrency is reduced and Qo S of replica is improved.At the same time,the framework adopts a master-slave working mode and uses a multi-master approach to avoid the occurrence of a single point.To address the problem of replica placement,through in-depth research on the IPFS cluster management tool—IPFS-Cluster,the related components are modified,and a node credit-based replica dynamic placement framework is proposed.Some parameters including hardware configuration,load,operating parameters,and churn times are used to calculate node credits and dynamically place replicas based on credits to improve the service quality and availability of replicas.Finally,it was verified by comparative experiment that the SVR-based visit heat prediction can achieve relatively good accuracy.And IDRM was tested from three angles of effectiveness,storage utilization,and access delay.The experiments proved that the IDRM system can effectively perform replica adjust and reduce response latency in high concurrent condition.