Processing of Mixed-Sensitivity Video Surveillance Streams on Hybrid Clouds

We consider a hybrid cloud model for video surveillance systems with mixed-sensitivity video streams. The hybrid cloud naturally addresses the issues on security by keeping sensitive data in the private cloud, and relieves seasonal workload by pushing computation to the elastic public cloud. Nevertheless, to enhance usability and reduce cost, it is desired to have a middleware that seamlessly integrates the two clouds and schedules the tasks effectively. We first present a stream processing model that is specifically designed for this hybrid cloud setting. Based on this model, we formalize the scheduling issue as an optimization problem that minimizes overall monetary cost to be incurred on the public cloud, with resource, security and Quality-of-Service (QoS) constraints. Our proposed scheduler exploits special properties of hybrid clouds for more effective solutions. Experiments through both large-scale simulations and prototype runs on Amazon EC2 show that the proposed approach is effective in outsourcing computational workload with overheads lower than other alternatives.