Abstract: Data availability and accessing performance in large scale storage network have been more and more important. Based on the in-band virtualization system for massive storage, virtual disk replicas are proposed to improve the data availability. The replicas use selective read scheduling, asynchronous update mechanism, and a special space-adjusting algorithm to optimize the data accessing performance. The experiments with virtual disk replicas show that, when the device number is adequate, the read-performance can be increased by 26%; even if one or two disks fail, the I/O operations can be processed correctly, and the read-performance is still over 10% better than the normal accessing without replicas.