This work package studies
the question of how to build a framework for real-time parallel and distributed processing of large-scale sensor fine-grained data in the cloud. This way, WP4 provides the cloud counterpart to the mobile framework that is built in WP3. This unique approach will give us the best of both worlds: low-cost analysis can be done quickly on the phone, but the system is still able to offload to the cloud when circumstances require it, e.g. in case of low battery level, or when the information from a single device is not enough. The end result is an open, cloud-based platform, based on open-source tools and frameworks.
State of the art solutions
with respect to large scale distributed processing are suitable for batch-type handling of large datasets and calculating transformations on these datasets in a distributed way. What we aim for, however, is large scale real time processing. This is an issue not resolved yet. It means a shift from a batch-oriented approach towards a streaming approach. We will create a scalable topology of interconnected sensor data streams and processes that analyse these streams. When the static structure of real-time analysis is in place, we will make the system dynamic and so it can be adapted to the context. Algorithms will be dynamic so the system does not have to be restarted when algorithms are updated.
Additionally, we will study
how all the information collected in the cloud can be usefully combined, integrated with external sources on the Internet, and fed back to the athletes, in order to increase their performance or adherence to a training scheme. Given the richness of sensor data, many scenarios are feasible here. Based on requirements from the application scientists, a small selection will have to be made that offers effective motivating feedback. A few illustrative examples are: comparing performance against other athletes or advising groups of athletes how to split up into subgroups with similar performance.