Cloud computing has become one of the most important trends in the field of Computer Science. It is a flexible, efficient and reliable method of allocating computing power to large-scale applications or services. A very common use of cloud computing is to schedule bags of tasks on machines and have them processed at a high rate of throughput. In scientific computing, scientists have the need to start from an initial bag of tasks, and to analyse the results at runtime, steering the execution of a particular simulation code in certain interesting directions by generating more tasks at runtime. The complexity of submitting tasks to the cloud combined with the multitude of scheduling algorithms that exist make up a very important research question for this execution model. The objective of this thesis is to modify the existing service, named Task Farming Service for the ConPaaS cloud platform such that it accepts new tasks at runtime. The aim is to use this platform to overcome the difficulties of interacting with the cloud and to develop a throughput-oriented model that allows the user to trade computation cost with a task processing rate rather than a static completion time.