Design And Implementation Of Multi-level Fault Tolerance Mechanism In Decentralized Storage System

Decentralized storage system based on P2 P network is an important branch of distributed storage.Traditional distributed storage systems are usually deployed centrally,with high reliability and easy maintenance,while decentralized storage systems built on P2 P networks face huge reliability introduced by low node reliability and maintenance uncertainty risk.Although reliability can be improved by adding redundancy or speeding up the repair rate,it also causes a significant increase in system cost.In view of the above problems,a layered fault-tolerant method is proposed,which stores the metadata and data in the system in layers according to volume and read-write frequency,and sets different redundancy and repair mechanisms for each layer of data to achieve control The overall cost of the system and the goal of maintaining higher reliability.Specifically,store stable and extremely small metadata on a blockchain with widely-reliable verification,make full use of the blockchain’s multi-copy feature to improve metadata reading performance,and avoid blocks The low transaction processing capacity of the chain becomes a system bottleneck;for other metadata that needs to be updated frequently and has a small volume,multiple copies are stored outside the blockchain,which maintains high reliability and effectively simplifies synchronization operations.And control the cost of redundancy;the larger volume of data should be stored as much as possible with a more cost-effective erasure coding method,and by dividing and conquering data and its digital signature redundancy,it also meets the needs of security and reliability.Realized the system prototype.Theoretical modeling and functional test results show that: under the selection of common fault-tolerant methods,the total redundancy of the prototype system(total data,metadata and redundancy / total data and metadata)and the redundancy of the fault-tolerant method Compared with the degree,the increase rate can be controlled within 4%;at the same time,the prototype system can timely identify and effectively repair the metadata and data lost due to the failure of various nodes to meet the basic fault tolerance requirements.