The present invention discloses methods and systems for reducing network latency. A first network device establishes a plurality of connections with a second network device. After that, determining non-congesting latency of each of the plurality of connections. Assigning a weighting to each of the plurality of connections. Decreasing the weighting of a connection when the performance of the connection deteriorated according to a first criteria. The first network device may perform weight decreasing again after a time interval. Last, sending data packets through the plurality of connections according to the weightings.