Research And Implementation Of Off-site Disaster Recovery Solution Based On Ceph Cloud Storage

With the rapid spread of the Internet in various industries,various business scales continue to grow,data is increasingly centralized,and individual data center backups can no longer meet the disaster tolerance needs of the business,such as the domestic hot Tmall double eleven activities.In foreign Black Friday,the explosive data request in a certain period of time,only relying on a single data center to operate,may cause the entire data center to collapse,so many cloud computing companies on the basis of single live to live off-site(Active-Active)disaster recovery solution.Off-site dual-active technology can greatly reduce backup costs,and automatically switch data centers to ensure the continuity of services in the event of a disaster.Therefore,researching off-site dual-active technologies is the key to improving data center disaster recovery capabilities.As a distributed storage system with reliable,automatic storage balancing and automatic data recovery,Ceph cloud storage is compatible with object storage,file storage and block storage.It also has a cluster monitoring mechanism Mon and a unique set of data.The distributed Crush algorithm is used to accurately locate the storage location of data objects.Crush data distribution algorithm is a pseudo-random algorithm used to manage the storage distribution and replication of data.It can meet the needs of Ceph cloud storage in large-scale distributed storage clusters to quickly and accurately calculate the storage location of data,and also in hardware failure.And make the smallest possible data migration with extended hardware storage scale.Based on the characteristics of Ceph cloud storage,this paper proposes a disaster recovery solution based on Ceph cloud storage for off-site dual-live online hot standby based on the in-depth study of Crush algorithm.The innovation of this scheme is as follows:(1)Propose an off-site dual-live online hot standby solution based on Ceph cloud storage.Two live-active Ceph clusters in different places provide services at the same time,and require Mons between clusters to communicate with each other through heartbeat detection,and update CRUSH MAP and OSD MAP information in real time.In addition,the administrator is the cluster of the designated party as the primary cluster.This designation is to add the primary cluster configuration to the MON configuration item.The default initial configuration of the configuration item of thecluster is disabled.The cluster of the designated party is the main cluster.When a brain splitting fault occurs,the primary cluster customer continues to serve externally.The non-primary cluster is aware of the unreachable gateway with the primary cluster through the heartbeat.Therefore,any IO operations for the non-primary cluster are It is in a blocked state until the network between the two clusters recovers.(2)Proposed double-write cluster data double write case plus inter-cluster consistency exclusive lock,can prevent the underlying data from being dirty.A lock is a means to prevent other transactions from accessing resources at the same time.A consistent exclusive lock can write the data at the same time when the remote client writes the data at the same time.The other party holds the write permission,and the other party writes the rejection.The queue waits for the exclusive lock release to regain resources.This ensures that multiple users can simultaneously manipulate data in the same underlying storage device without data inconsistency,so that off-site dual-active technology can be better applied.(3)In the case of dual-active,the inter-cluster consistency shared lock is used to isolate the read and write operations of shared resource data in the same time.When two clients access the resource at the same time,one party initiates a write request to the resource,and the other party initiates a read request to the same resource,and the heartbeat detects that the read request party first acquires the consistency shared lock,then the write request operation will Blocked and waiting in the queue for fast access to data.By using the fast location location feature of the Crush algorithm,it can not only ensure the user's fast read request,but also ensure the accuracy of data reading.In the paper,the simulation of the data storage and reading test of the dual-live clusters with different clients is carried out by using the simulation of the off-site Ceph active-active cluster.The experimental results show that the proposed off-site dual-live online hot standby solution can not only improve the data storage.The security and read accuracy,and in the case of simulating the main cluster disaster recovery,can ensure business continuity,achieve the goal of no data loss with RPO equal to zero and the shortest data recovery time point of RTO.