STP - Root Bridge

[muca]

Hi,

I read that although the root bridge election process is automatic you should manualy select a primary root bridge and a secondary root bridge because you could have an access switch acting as the root instead of a more powerful equipment.

Is there any rule about choosing the primary root bridge? (For instance the most powerful switch in the Network)?

Thanks

[Francois Tallet]

Well, in fact, processing bpdus is more expensive in term of CPU than sending them. So the root (which generally does not block port and thus almost only sends bpdus) is not necessarily the most impacted.

The most important is to select a root that has a central position in your network. Here, I mean central with regard to the data traffic exchanged. The path between two distribution switches will likely go through the root, even if they are directly connected. So the root must be strong in term of switching capabilities more than in term of CPU power. Generally, this goes in pair, so you can safely summarize that as: you'd rather have a powerful bridge as root of your network;-) But its location is also very important.

Regards,

Francois

[spvaidya]

Hi Francois,

Referring your views has some doubts. If i've a three tier architecture (Core,Distri and access with 6500 as core, 3560 as distri and 29XX as access) then 6500 will be more powerful and 3500 is more centrally placed.

Which will be the best layer to put up root bridge?

Regards,

Shashank

[jpl861]

per your design, broadcast should not traverse the core layer. broadcast should stop in the distribution layer.

So the layer 2 part or your network stays from the distribution layer to your access layer. Connection between core and distribution should be layer 3.

So in your design, you can set your 3560 switches as your root bridge.

This is just a suggestion.

[Francois Tallet]

I guess it's a question of terminology, but why do you consider your 6500 in the core if the 3560 is more centrally placed;-)?

Or maybe I did not defined "central" correctly. I did not mean "in the middle" of the network diagram. Be careful that the network diagram is generally representing the core at the top (ie at the edge of the diagram), which may be misleading. Reorganize you network diagram by putting the distribution around the core bridges, and the core should look more central.

If you have the typical hierarchical design you probably don't want the root to be in your distribution. As a result, two directly connected bridges in your core might communicate together through your distribution switch. That basically means that your distribution switch is part of the core. I would assume you rather want a cat6k to be the root.

Let me just describe another way of seeing this. At L2, there is only a single topology for a given vlan. This topology (the tree) is centered on the root. That means that the path from anywhere in the vlan to the root is optimal. The root is the only node in the network that has this property. So consider who is initiating traffic in your network, and where it is going. If you have lots of clients spread across your network accessing a default gateway or some servers, then it makes sense for these gateways or servers to be as close as possible to the root. This way, the traffic from the client will go through as few links as possible.

I'm rather a visual guy, it's difficult to discuss those things without a whiteboard;-) Let me know if it helps.

Regards,

Francois

[spvaidya]

Francois, sincere thanks for taking that efforts. Your explanation is of great help.

Think the decision based on the traffic pattern is of more importance to decide the root in the network. correct me if am wrong.

1) If traffic is server centric (with servers sitting near to core - DC), root bridge near to servers is a good design, AND

2) If traffic pattern is both, talks to servers as well as between the VLANs ( Access), then root bridge can be put at the distribution layer as well ( right ?)

Regards

Shashank

[Francois Tallet]

Hi Shashank,

Well, if you had a full mesh between your distributions, I would agree on 2). But generally, the cabling itself is biased toward a core centric network. Because I really need a diagram, I've created a small pdf with a typical hierarchical network with the root at the distribution and the root at the core. Bold links are the representing the tree (black dots blocked ports).

You can see that if you distribution is root, you've not really enhanced the communication between the clients connected to the access. The is because you need to go through the core to communicate between distribution switches anyway. Basically, distribution switches don't have enough connectivity the one to the other. If you had a link R-5 for instance, that would really help. With this diagram, you have pushed core bridge 2 in a less central situation. A (generally) high capacity link 1-2 is now blocking. Also, hosts connected to access switches 9 and 10 need to go through 5-1-R to reach servers attached to core 2.

So basically, you have not improved connectivity between your end devices, but you have made connectivity to the services in your core worse.

If you were adding links between the distribution switches, then it would of course be worth. But in that case, you would in fact merge your distribution switches into a larger core. That's again a terminology issue at that stage;-)

Regards,

Francois