One solution is the usage of the rapid spanning tree protocol (RSTP -> 2004 edition of IEEE 802.1d), which actively informs neighbouring switches about port state updates for a coordinated selection of the new root bridge and the respective selection of root bridge ports.
Both protocols use path cost values as the first metric to determine the best path to the root bridge.
The default value of the root path cost is derived from the media speed of the root port interface.
If I am not mistaken, the original IEEE port cost value definition was as follows:
- 10 Mbps -> 100
- 100 Mbps -> 10
- 1 Gbps -> 1
- which may also lead to the following assignments: 10 Gbps / 100 Gbps / 1 Tbps -> 1
- 10 Mbps -> 100
- 100 Mbps -> 19
- 1 Gbps -> 4
- 10 Gbps ->2
- which may also lead to the following assignments:
- 100 Gbps / 1 Tbps -> 1
Is this latter assumption true and implemented this way in modern switches ? OR is the 100 Gbps / 1 Tbps issue not yet implemented by Cisco, Juniper (just to name a view) ?
Besides this cost definition problem, there are other issues of worldwide Ethernet based carrier networks using some sort of spanning tree. For instance, per-VLAN spanning tree or multiple spanning tree (MST) configurations help to subdivide the spanning tree regions in order to mitigate its slow performance. Furthermore, a single root bridge for a global Ethernet network does not seem to be very sensible anyway.
Please let us know, when you find articles about such mitigation strategies or even new standards, which solve the loop avoidance issue for single owned or inter-carrier Ethernet formations.
