Vehicle GPS data is an essential "raw" material for a broad range of applications such as traffic management and control, routing, and navigation. To become useful, the data has to be related to the underlying road network by means of map matching algorithms, which are often computationally expensive. In addition, GPS data is not accurate and often needs to be cleaned to remove erroneous observations. In this paper, we describe how map matching can be run on IBM's System S, which provides a platform to run stream processing applications in a scalable manner. We show how various features of System S, including a component based programming model, data pipelining and parallelization of computation, help us to scale the map-matching and data cleaning processes, both as the rate of incoming GPS data increases and as the size of the underlying road network increases. We provide results of performance evaluations, where we show our system can match GPS data arriving at a rate of 1 million points per second onto a map with 1 billion links. © 2010 IEEE.