Disk I/O may be the bottleneck in computer systems because of the mechanism characters in disk system. In order to tune the system performance effectively, the collection of the disk I/O workload characteristics will be the essential step for the performance optimization. Compared with other I/O characteristics collection methods, this paper presents an on-line I/O characteristics analysis method based on Xen 3.0 virtual machine system. In virtual machine environments, this method can be transparent to the unmodified operating system. With this method, several essential I/O characteristics metrics can be collected, such as disk I/O block size, I/O latency, I/O arrival interval, I/O spatial locality statistics, I/O time locality statistics and the I/O operation hotspot distribution. Through the testing and analysis, this method is found to have little system overhead and impact on the application system I/O performance. In addition, the I/O characteristics analysis results are demonstrated on the large file copy workload and the Filebench benchmark workload generated by the filemicro and varmail personalities.