As a new type of storage media, solid state drive (SSD) is widely used in virtualization environment. SSD is usually used as the read and write cache of the virtual machine (VM) storage to improve the disk I/O performance of the VMs. Existing SSD caching schemes mostly focus on the capacity planning of the SSD cache and use metrics such as cache hit rate to evaluate the effect of SSD cache allocation. Since they do not consider the limitations of service capabilities of SSD, which may lead to the contention of cache resource and the performance degradation and violations among VMs, they are not suitable to be used with some typical distributed applications. This article proposes a self-adaptive SSD caching system for multiobjective optimization in virtualization environment, to reduce the resource contention, and take the limitations of service capabilities of SSD into consideration. With the help of closed loop adaption, it can dynamically detect the status of VMs and applications. Moreover, it continuously detects the contention of SSD cache, generates the migration plan using clustering algorithm and decides the timing and order of the VM migrations according to the capabilities of SSD as well as the characteristics and requirements of applications. The evaluation shows that when facing the scenarios of using some typical distributed applications, the contention of SSD cache resource is reduced and the requirements of applications are considered, which lead to the improvement of performance and reliability of applications. For Hadoop applications, the execution time of jobs is reduced by 25% on average, and the throughput for I/O sensitive applications is improved by 39%. For ZooKeeper applications, the service outage caused by the single point of fault of the hypervisor can be handled at the cost of less than 5% of performance degradation.