| 摘要 |
A novel packet filtering and processing mechanism solves bandwidth estimation problems related to cross-traffic, host limitations, and fabric-specific issues in order to accurately estimate bottleneck bandwidth in fluctuating environments such as home networks. In particular, the process controls the number of packet train experiments, the size of trains and packets within trains, and other factors to eliminate errors due to cross-traffic, network capture, receiver CPU overload, interrupt moderation, and other networking factors. In an embodiment of the invention, a history window is applied to estimates derived from individual trains to ensure that data used to make a bottleneck bandwidth estimate was taken at a time when the network is relatively stable.
|
