Tree-based overlay networks for scalable applications
Dorian C. Arnold, Gary Pack, Barton P. Miller · 2006
The increasing availability of high-performance com-puting systems with thousands, tens of thousands, and even hundreds of thousands of computational nodes is driving the demand for programming models and in-frastructures that allow effective use of such large-scale environments. Tree-based Overlay Networks (TBŌNs) have proven to provide such a model for distributed tools like performance profilers, parallel debuggers, sys-tem monitors and system administration tools. We demonstrate that the extensibility and flexibil-ity of the TBŌN distributed computing model, along with its performance characteristics, make it surpris-ingly general, particularly for applications outside the tool domain. We describe many interesting applica-tions and commonly-used algorithms for which TBŌNs are well-suited and provide a new (non-tool) case study, a distributed implementation of the mean-shift algo-rithm commonly used in computer vision to delineate arbitrarily shaped clusters in complex, multi-modal fea-ture spaces. 1.