| 摘要 |
<p>An objective of the present invention is to improve overall processing speed by reducing the risk that a prescribed number of instances of target data will not be collected and suppressing variation in processing time among CPUs. Provided is a parallel distributed processing system which distributes and processes data to be processed upon a plurality of computation devices, comprising: a generating unit which generates a classifier which classifies target data from the data to be processed which is distributed upon the plurality of computation devices; a collection unit which collects, from each of the computation devices, the target data, processed data which is processed when the target data is classified among the data to be processed which is distributed, and unprocessed data which is other than the processed data among the data to be processed which is distributed; a distribution unit which distributes the data to be processed to the plurality of computation devices, aggregates the processed data and the unprocessed data which is collected from each of the computation devices on the basis of performance time when each of the computation devices has processed the processed data, and separates the collected processed data and unprocessed data as the data to be processed on the basis of estimated time which equalizes the time to classify the target data upon the computation devices; and a predict unit which predicts the estimated time which equalizes the time to classify the target data in the aggregated data to be processed upon the computation devices. The system repeatedly redistributes the separated data to be processed until a condition is met.</p> |
