Abstract | In a future where multiple cloud providers coexist, end users will have the option to select machines from multiple clouds according to the type and cost of offered resources that best suit their application. The resulting distributed infrastructures, which we refer to as sky-computing systems, will be inherently disconnected and heterogeneous with respect to performance. In order to efficiently scale-out an embarrassingly parallel application in this environment, existing technologies need to be combined and extended. This demonstration illustrates and end-to-end approach to sky computing. In order to achieve scalable management and performance, several technologies and techniques are combined, namely, an IaaS cloud toolkit (Nimbus) to create virtual machines (VMs) on demand, a virtual networking middleware (ViNe) to connect VMs, the MapReduce framework (Hadoop) for parallel fault-tolerant execution of an unmodified application, and a skewed task distribution technique to deal with resource imbalance. A bioinformatics application (BLAST) is used to demonstrate the scalability of this type of sky-computing environment. |