Abstract
We consider the challenge of incorporating existing and legacy application models into real-time geocomputation infrastructures. Several steps of translation and pre-processing are required before data can be consumed by models that have so far assumed a statically derived data set or must use coarse information due to computational constraints. A systematic procedure to provide dynamic data to legacy models saves software rewrite costs and enables the infrastructure to minimize communication costs of sending raw data to the back-end models. We describe our system that filters and aggregates real-time data before it sends it to back-end models. We focus on the location (within the networking infrastructure) of the computation and communication steps required for the pre-processing of raw data. We convert the pre-processing needs of a specific class of legacy models, whose spatio-temporal data requirements can be identified in a process-flow graph, into processing modules distributed in our infrastructure. We choose a placement of the processing modules directed at minimizing communication costs.
