Bridging the Gap between High-Performance, Cloud and Service-Oriented Computing

Modern large-scale computing systems are often naturally assumed to be based on the cloud provisioning paradigm. While this model is very flexible and often more cost effective than the actual ownership of physical servers, it also comes at a price, i.e., abstraction by virtualization. Even today, there are many fields of application in which any kind of performance degradation would be considered harmful. These types of applications are what we consider to be the core of classical high-performance computing. Bridging the gap between this low level performance optimization oriented execution model, higher level but more flexible cloud-based Infrastructure-as-a-Service provisioning and finally service-oriented execution models on the same physical hardware is a very challenging task. It requires tailor-made solutions to connect and combine the various independent software layers and abstraction stacks into a single manageable platform. In this paper we present our approach for the integration of all three tiers into one large scale hardware and software framework. Based on our existing solution for the simultaneous operation of cluster and cloud infrastructure on the same physical hardware, we provide an outlook on our current work-in-progress and vision for the coupling of two remote locations into a single distributed system for service-oriented applications.