Security is a major challenge in Cloud Computing. In this paper, we propose an Online Smart Disguise Framework (OSDF). OSDF employs dynamic, proactive, real-time moving-target defense against cloud attacks. OSDF relies on two main pillars. The first, is a behavior obscuring module that frequently live-migrates virtual machines (VMs) between heterogeneously configured compute nodes to avoid co-residency and virtualization based attacks. The second module limits attack dispersion between same-host VMs by migrating maliciously behaving VMs to remote isolated compute node acting as a quarantine zone. The second module is guided by a smart intrusion detection system that monitors the VM system calls searching for suspicious activities. To evaluate OSDF efficiency and effectiveness on limiting attack dispersion, we devised the vulnerable, exposed, attacked, recovered model based on the susceptible, exposed, infected, recovered (SEIR) epidemic model. The SEIR model is an epidemiological model commonly used to investigate disease dispersion on cooperative communities. The implementation of OSDF is tested on OpenStack private cloud. Simulation results show the effectiveness of OSDF MTD approach in decreasing the number of attacked VMs even for fast-spreading worms. Furthermore, NAS Parallel Benchmark is used to evaluate OSDF efficiency for cloud-hosted VMs running both stateful and stateless applications.