In today’s ever computerized society, Cloud Data Centers are packed with numerous online services to promptly respond to users and provide services on demand. In such complex environments, guaranteeing throughput of Virtual Machines (VMs) is crucial to minimize performance degradation for all applications. vmBBThrPred, our novel approach in this work, is an application-oblivious approach to predict performance of virtualized applications based on only basic Hypervisor level metrics. vmBBThrPred is different from other approaches in the literature that usually either inject monitoring codes to VMs or use peripheral devices to directly report their actual throughput. vmBBThrPred, instead, uses sensitivity values of VMs to cloud resources (CPU, Mem, and Disk) to predict their throughput under various working scenarios (free or under contention); sensitivity values are calculated by vmBBProfiler that also uses only Hypervisor level metrics. We used a variety of resource intensive benchmarks to gauge efficiency of our approach in our VMware-vSphere based private cloud. Results proved accuracy of 95 % (on average) for predicting throughput of 12 benchmarks over 1200 h of operation.