Challenges in real-time virtualization and predictable cloud computing

The cloud computing paradigm and its key enabling technology, i.e., virtualization, has been studied and advanced in the past decade for the mainstream distributed systems domain. The reduction of operation and maintenance costs but outsourcing them to data centres has become a revolution in the way that users and companies operate on a daily basis for interacting, exchanging information, and saving data. This has been enabled by the virtualization technology can provide different customized execution environments over a same physical node. In mainstream cloud computing, the hardware is typically buried under the software virtualization layers, and applications do not by pass it. However, real-time virtualization technology must provide temporal and spatial isolation among applications as well as access to the hardware resources. Virtualization is, however, still a strong challenge to be applied in real-time systems. The temporal properties of real-time applications have to be preserved. In this paper, authors describe the implications of merging real-time and cloud computing as a way to understand the challenges that this brings about. We explain the details of VM architectures and the role of hypervisors, providing also a conceptual mapping between real-time and cloud computing terminology. Also, we explain the issues on scheduling for cloud and the specific network problems. Finally, we provide a set of directions enumerating some the most challenging problems ahead of us.