Authors: Leilani Battle, Remco Chang, Michael Stonebraker
Abstract: With many current visualization systems, users must manually throw data away until it fits in memory, before they can visualize it. We propose instead to expose this resource-latency tradeoff to the user directly, by allowing the user to specify resource constraints and have the system adjust automatically. In this paper, we present ForeCache, an exploration system that visualizes aggregated views of datasets stored in a DBMS. We implemented a number of server-side techniques in ForeCache for prefetching small subsets of aggregated data (i.e. chunks) for fast visualization of large datasets. Our techniques leverage locality in the user’s exploratory behavior, and improve upon existing techniques in two ways. First, instead of pre-computing all data chunks in advance, we reduce storage requirements by only pre-computing a subset of chunks in advance, and computing the remaining chunks at runtime as needed. Second, we balance runtime computation costs by predictively building and caching new chunks in anticipation of the user’s needs.