A topology-aware adaptive deployment framework for elastic applications

In Distributed Cloud Computing, applications are deployed over thousands of geographically distributed cloud sites. This new deployment approach promises not only improved application's quality of service but enables deploying network-critical applications otherwise not possible. A previously settled, static allocation of enough resources at each site is expensive. Adapting resource allocations during application lifetime could dramatically reduce expenses. They are triggered by complex algorithms as a reaction of changes in measured performance data. However, such adaptation algorithms and their performance data depends on specific application's requirements, constrains, and on optimization goals. Some interactive applications need a low packet round trip time. Other streaming applications need a high data rate to the user. Besides these individual characteristics of an adaptation, similarities exists in the management of such a geographical distribution: Applications and their components are deployed within Virtual Machines. The components have to find and communicate with each other. Motivated by this, we present a framework taking care of necessary common functionality while been highly customizable to support a wide range of adaptation. Additionally, integrated adaptations can utilize combined application- and infrastructure- level data and are also able to reconfigure the application and the infrastructure. Finally, a steering-system supports state-ful and multi-tier applications to be deployed in an elastic and dynamic way.