With growing number of applications using web based transactions and services, development of tools and techniques for analyzing large website usage data for enabling deeper insights into usage trends and hidden patterns is a major requirement today. We propose a solution approach for interactive visual analysis of website usage, which integrates usage modeling and visual rendering. Our approach not only makes the approach scalable to large data by reducing the time for visual rendering, but also enables easier interaction for subsequent analysis by visually presenting responses to queries in the global context of historic usage behavior. As the first step, we apply a fuzzy clustering technique to web log data to obtain a usage model and image it through a two-view display showing a point cloud rendering of clustered sessions and the website page hierarchy. Since web usage data is high-dimensional and non-Euclidean, we combine two dimensionality reduction techniques, namely Multidimensional Scaling and Sammon mapping for transforming the usage session data into displayable primitives. To demonstrate the effectiveness of our approach, we have developed a prototype system and performed experiments using large datasets which supports (1) clickstream visualization to identify macro level trends in real time, (2) decision on when web site restructuring is needed based on a proposed cost model while interactively rearranging nodes in the hierarchical display, and (3) visual depiction of noise for intuitively deciding upon when to carry out the expensive process of reclustering data. We also illustrate the effectiveness of the proposed approach using some benchmark datasets.