The problem of task scheduling is as old as grid computing and has been solved many times under different conditions. In this paper, we consider an algorithm for the task scheduling in the heterogeneous computer network of the geographical information system O-GIS, which includes dedicated data servers, central computing nodes, and end-user clients. Since some client devices in our network are also computing nodes, not only available computation power is considered as the influential factor, but also various network quality in incoming and outgoing communications of each node on computing grid. Using the specificity of raster computation, the original problem is formulated and its reducibility to the problem of mixed linear integer programming is shown. We also propose a method of deriving a feasible near-optimal solution without solving an NP-hard problem, which is used in O-GIS since we do not need an exact one as our network is not stable enough for the optimal solution to be significantly better in real-life cases.