Route control method and route setting device
The route control method and the route setting device according to the present invention is configured such that a learning unit is prepared for each path, a function form of the learning unit of the path in forwarding a packet is adjusted by having traffic information (packet arrival rate, queue length, and the like) and congestion information (congested or not congested) of this path as teacher data, and when this path is congested, changing to a path with an output of no congestion among the learning units is executed. Thus, since the learning unit optimizes the function form, changing to an alternative path can be immediately executed even when the congestion incidentally occurs on the path.
1. A route control method comprising:
preparing learning units for each of a plurality of paths having a source point router and a destination point router as an identical router in a communication network including a plurality of routers;
accumulating, in the learning units prepared in each of the plurality of paths, traffic information of all of the plurality of routers in the communication network and congestion information of the each of the plurality of paths, wherein the traffic information includes a queue length, a number of packet arrivals, and a packet discard rate of each of the plurality of routers;
performing machine learning for the each of the plurality of paths using the accumulated traffic information and the accumulated congestion information as teacher data, wherein the machine learning is performed by the learning units;
performing congestion prediction based on results of the machine learning for the current traffic information, the congestion prediction being performed by each of all the learning units; and
setting one path predicted not to be congested for a path from the source point router to the destination point router.
2. The route control method according to claim 1 , wherein a shortest path is selected among the plurality of paths to start the machine learning, and the machine learning and the congestion prediction are repeated for a predetermined time.
3. The route control method according to claim 2 , wherein
a path other than the one path predicted not to be congested for the predetermined time with a predetermined probability is set for a path from the source point router to the destination point router.
4. A route setting device comprising
learning units prepared for each of a plurality of paths having a source point router and a destination point router as an identical router in a communication network including a plurality of routers, wherein
the learning units prepared in each the plurality of paths:
accumulates traffic information of all of the plurality of routers in the communication network, and congestion information of the each of the plurality of paths, wherein the traffic information includes a queue length, the number of packet arrivals, and a packet discard rate of each of the plurality of routers, and
performs machine learning for the each of the plurality of paths using the accumulated traffic information and the accumulated congestion information as teacher data, and
each of all the learning units:
performs congestion prediction based on results of the machine learning for the current traffic information, and
sets one path predicted not to be congested for a path from the source point router to the destination point router.
5. The route setting device according to claim 4 , wherein
the learning units select a shortest path among the plurality of paths to start the machine learning, and
the machine learning and the congestion prediction are repeated for a predetermined time.
6. The route setting device according to claim 5 , wherein
the learning units set a path other than the one path predicted not to be congested for the predetermined time with a predetermined probability fora path from the source point router to the destination point router.