Using Virtual Private Networks for Reliable VM Based HPC Systems

The emergence of cloud computing and virtualization is creating a revolution in the design and deployment of large-scale distributed systems. The powerful capabilities of these technologies can be used to create highly reliable virtual high performance computing systems. However, a drawback of these technologies is the lack of information and limited control over the network infrastructure interconnecting geographically distributed virtual machine instances. An attacker with access to the underlying infrastructure can easily monitor or disrupt the virtual cluster. Virtual LANs (VLANs) and firewalls can be used to create private networks for virtual clusters. These approaches, however, are time-consuming to configure, limit the geographic zones in which a virtual cluster can operate, and complicate the management and migration of virtual nodes within a cluster. Moreover, proactive and reactive fault tolerance strategies that rely on the live migration of virtual machine images require a uniform network address space to facilitate the quick migration and restart of VM images running parallel applications. In this paper, we present a new approach for reliable high performance computing built on virtual machines that is based on a P2P VPN that provides a secure and uniform address space for parallel applications running within a set of VM images. We demonstrate the functionality and performance of our approach using OMEN and a set of MPI performance benchmarks.