| 摘要 |
In a butterfly network, a number of switches are set to provide two paths that are independent of each other, from a first switch to a second switch, and from the first switch to a third switch respectively. Identification of switches to be set from among all switches in the butterfly network depends on the locations of the first switch, the second switch and the third switch relative to one another. The to-be-set switches are determined by starting with the first switch as a preceding switch, identifying the next switch for a path by simply changing the level number (e.g. incrementing the level number) of a preceding switch in the path, and by changing a bit of the row number of the preceding switch (e.g. by replacing the alpha-th bit with a corresponding bit from the destination switch's row number), and repeating such acts with the just-identified switch as a preceding switch. The direction of the path is reversed on reaching a last level or a last row of the network. Such addressing techniques identify all switches that need to be used to form two node disjoint paths from the first switch to the second and third switches. The two paths can be used to redundantly couple a source switch to a destination switch, for load balancing, for fault tolerance, or for multicasting.
|
