Towards Efficient and Precise Queries over Ten Million Asteroid Trajectory Models

We consider the problem of finding known asteroids in a given region of space and time. Each asteroid’s trajectory is known precisely at any given point in time, but evaluation of the model describing its location is prohibitively expensive at query time. We propose a framework for sampling the base model and indexing the resulting “observations.” Our system includes 11 million simulated asteroids, and our requirements include 5 milliarcsecond accuracy and 15 second query response time. We evaluate the effect of various sampling and indexing alternatives in this approach and propose two exemplar solutions: a faster but space-intensive interpolation-based method, or a slower method that relies on bounding regions tailored to the object’s behavior but requires fewer sampled points. We implement all solutions in a relational database and evaluate them on a full-scale dataset of known asteroid trajectories.