Long running multi-user server system may encounter frequent errors resulting in running disruptions due to its complexity of program, operating environments and user operations. This poses the need of self-recovery of system. Rollback and checkpoint scheme is a popular self-recovery strategy in current research and application, but has no obvious effects in multi-user system. In this paper, a VMM-based self-recovery system named VMSRS (virtual machine monitor-self recovery of service program) is designed according to the characteristics of multi-user server programs. The main idea of VMSRS is regarding VMM as major component of recovery, taking advantage of VM as independent underlying system and hardware resource monitor, and strictly maintaining the consistency and security of user data and atomicity of data operation. As an improved SRS (self recovery of service program), VMSRS controls errors to avert affecting normal users in case of system crash instead of committing rollback, allowing users and servers to proceed as if no crash happens. Rollback is avoided by taking advantage of self-cleansing mechanism of system and VMSRS. The issues addressed by VMSRS design include crash suppression module, demand driven restoration module, monitor module, and storage management module. The experiment results from analyzing basic function, basic performance and integral function validate that VMSRS can provide favorable security and consistency of user data while guaranteeing performance and committing no rollback. It recovers multi-thread programs excellently with no limit to threads. Meanwhile, this exploratory study also takes part in current research of self-recovery system utilizing virtualization technology.