On optimal mapping of visualization pipeline onto linear arrangement of network nodes

This paper discusses algorithmic and implementation issues of optimally mapping a visualization pipeline onto a linear arrangement of wide-area network nodes to minimize the total delay. The first network node typically is a data source, the last node could be a display device ranging from a personal computer to a powerwall, and each intermediate node could be a workstation or computational cluster. This mapping scheme appropriately distributes the filtering, geometry generation, rendering, and display modules of the visualization pipeline to the linear arrangement of network nodes to make efficient use of the computing resources at end nodes and also the network bandwidth between them. A regression based network daemon is developed to measure the available bandwidth on a transport link. We present an analytical formulation of this problem by taking into account the computational power of nodes, the bandwidths between them, and the sizes of messages exchanged between visualization modules. We propose a polynomial-time optimal algorithm that uses the dynamic programming method to compute the mapping with the minimum total delay. An OpenGL-based remote visualization system is implemented and deployed at three geographically distributed nodes for preliminary experiments.