Mendel Rosenblum, and
Computer Science Department
In Proceedings of the Sixth Annual Symposium on High-Performance Computer Architecture
We demonstrate the effectiveness of this parallel analysis and visualization system with a case study. We developed two visualizations within Rivet to study the Argus parallel rendering library, focusing on the memory system and process scheduling activity of Argus respectively. Using these visualizations, we uncovered several unexpected interactions between Argus and the underlying operating system. The results of the analysis led to changes that greatly improved its performance and scalability. Argus had previously been unable to scale beyond 26 processors; after analysis and modification, it achieved linear speedup up to 45 processors.
PostScript (3 MB gzipped, 9 MB uncompressed), Color PostScript (3 MB gzipped, 11 MB uncompressed), PowerPoint presentation (1 MB).