Linux-based Asynchronous Remote Volume Replication System

Today more and more applications want to keep data functioning 24 hours a day and seven days a week, however, unplanned events would lead to data loss and seriously damage business continued operations, which cause tremendous losses to companies. A real-time disaster tolerance solution is required for these environments; it can minimize or eliminate data loss with the help of remote replication, and recover quickly with real-time backup data. Efficient data replication system has become the key part of disaster tolerant system, but its complexities are surprisingly poorly understood, and possible optimizations are infrequently taken advantage of, so data is less well protected than it ought to be.By the in-depth study and analyze about existing data replication systems and relevant technologies, we design and implement a replication system called ARVRL (Linux Asynchronous Remote Volume Replicator for Linux), which describes an optimization design of asynchronous remote replication. ARVRL provides a real-time remote replication of local volume over any IP network for maximum business continuity, it's independent on specific network devices; it lowers the response time of replication process by asynchronous mode and decreases the workload of network with write coalescing; it can supplies any volume devices.The main contribution in the thesis can be summarized as follows:(1) introduces the related research on remote replication, include the design goals and design modes, then discuss the way to maintain data consistency.(2) Explains the basic concepts and designing structures in ARVRL system, and figures out the work flows and state machine.(3) Based on data flows of writing requests, the replication process is divided into several individual parts. This division minimizes the possibility of visiting critical resources and makes the architecture of the whole system more clear.(4) Designs an remote replication virtual block device on Linux, which can deal with the writing requests and replicate it to remote site.(5) Designs commit unit propagation with write coalescing, which decreases the workload of network, and introduces a formula which dynamically adjusts the size of coalition period to balance the amount of data transmitted and data loss.Finally, we report on correctness and performance validation of ARVRL, which prove that it supplies users with an economical and applicable kind of data replication system, there is also certain theoretical value in optimization design of ARVRL.