In cloud environment, most services are provided by virtual machines (VMs). Providing storage quality of service (QoS) for VMs is essential to user experiences while challenging. It first requires an accurate estimate and description of VM requirements, however, people usually describe this via rules of thumb. The problems are exacerbated by the diversity and special characteristics of VMs in a computing environment. This paper chooses Virtual Desktop Infrastructure (VDI), a prevalent and complicated VM application, to characterize QoS requirements of VMs and to guarantee QoS with minimal required resources. We create a model to describe QoS requirements of VDI. We have collected real VDI traces from HP to validate the correctness of the model. Then we generate QoS requirements of VDI and determine bottlenecks. Based on this, we can tell what minimum capability a storage appliance needs in order to satisfy a given VDI configuration and QoS requirements. By comparing with industry experience, we validate our model. And our model can describe more fine-grained VM requirements varying with time and virtual disk types, and provide more confidence on sizing storage for VDI as well.