Research On Multi-datacenter Distributed Block-level Storage

With the more and more widely appliance of datacenters, the requirements of data sharing between datacenters are becoming more and more imperative. Storage systems of datacenters have different architecture and data access interfaces, this creates difficulties for unified management, efficient storage, data retrieval and update; Storage resources of multi-datacenters are widely distributed with narrow bandwidth and low transmission rate, this makes great impact on performance of data access and applications.In this paper, we conduct research on problems mentioned above, work includes:According to characteristics of WAN-based storage resource management framework, we propose a multi-level across datacenters storage resources management framework. By using block-level storage we can effectively reduce the logical storage hierarchy, shorten I/0 requests route and improve system performance. Multi-level storage resource management helps to divide the administration load to different management layer, reduce administrative burdens and lower the difficulty of system design and implementation.For WAN and data sharing across the datacenter, we propose a across datacenter cache mechanism. The client cache can effectively enhance the dada access performance, volume-level atomic update mechanism can quickly update the data while maintaining the data consistency.We have designed and implemented a multi-datacenter distributed block storage system called Federal Volume Manager(FVM). Then we test the performance of FVM. Test results show that FVM has relatively high data access performance on WAN.