We present the results of the first online measurement in ATLAS of the LHC beam position and size at sqrt(s) = 900 GeV in 2009 and sqrt(s) = 7 TeV in spring 2010. A dedicated algorithm, implemented in the ATLAS Level 2 Trigger, takes fully reconstructed tracks in the Inner Detector as input to a fast vertex fitter in order to reconstruct vertices on an event-by-event basis. The three-dimensional distribution of primary vertices carries information of the LHC luminous region at the ATLAS interaction point and is used to extract its position, size and tilt angles. The luminous region parameters are monitored in real-time and used for feedback to the LHC. With this method, we observe changes in the transverse centroid position that mirror IP-orbit drifts, as well as longitudinal shifts arising from RF phase changes. Also, variations in the transverse widths, and an expected increase in the longitudinal spot size over the course of a fill were seen. In addition, the measured beam spot is used to track significant changes in the accelerator, which can then be redistributed to the High-Level Trigger for use by trigger algorithms that depend on the precise knowledge of impact parameter or decay length, such as b-tagging. We will present the techniques developed to allow such real-time configuration changes on the High-Level Trigger farm of currently 810 processing nodes in a way that does not disrupt data taking or incur deadtime, while ensuring a consistent and reproducible configuration across the farm. Lastly, by counting the primary vertices online, we use this same algorithm to provide online monitoring of the instantaneous luminosity of the accelerator. The beam position measurements presented here were available in real-time and used to provide feedback to the LHC operators for beam adjustments during the first LHC runs. Show Partial Abstract