摘要 |
An improved method for dividing and distributing the work of an arbitrary algorithm, having a predetermined stopping condition, for processing by multiple computer systems. A scheduler computer system accesses a representation of a plurality of work units, structured as a directed graph of dependent tasks, then transforms that graph into a weighted graph in which the weights indicate a preferred path or order of traversal of the graph, in turn indicating a preferred order for work units to be executed to reduce the impact of inter-work unit dependencies. The scheduler computer system then assigns work units to one or more worker computer systems, taking into account the preferred order. |