The visualization groups at LBL and SDSC teamed up to study the performance of commonly used visualization algorithms for streamlining computing on Dash. By modifying the algorithm to use disk-based cache, the group saw an increase of two to five fold over traditional methods. Specifically, the group found that the fast storage times of SSDs led to significant gains over local hard drives in some cases. The study will be presented at the "IEEE Symposium on Large-Scale Data Analysis and Visualization" at Providence, RI, USA on October 23–24, 2011.
The goal of the MOPS project is to identify asteroids and other transient near-earth objects from images that will be collected by the Large Synoptic Survey Telescope (LSST). Working with synthetic data, we are collaborating with the LSST software team to develop fast, efficient algorithms for several stages of the data processing pipeline. Through a combination of serial optimizations and implementation of thread-level parallelism, the time to solution for two key steps of the process – removal of subset tracks and linking togther of “tracklets” to form complete tracks – has already been reduced by 4–20x. Due to the need for large, shared memory, the software is ideally suited for the vSMP nodes on Dash and Gordon.