Jump to content

Hi,

 

I've started to look more into the whole spine-leaf topology topic but I'm not sure I really get it on a deeper technical level. The most confusing aspect stems from all the "No STP" talk.

 

For the sake of simplicity let's just say I have one server, two leaf switches and two spine switches. So now I connect both NICs of the server with both leaf switches and both leaf switches with both spine switches. Neither spine nor leaf switches are directly connected to each other. (Leaf - Leaf or Spine - Spine)

 

So far so good. What's confusing me the most is all this talk about implenting this via L3 or L2. Does Layer 3 implementation mean that every uplink from Leaf to Spine has it's own network? For example:

 

- Leaf1 Port1 10.0.0.1/24 to Spine1 Port1 10.0.0.2/24
- Leaf1 Port2 10.0.1.1/24 to Spine2 Port1 10.0.1.2/24
- Leaf2 Port1 10.0.2.1/24 to Spine1 Port2 10.0.2.2/24
- Leaf2 Port2 10.0.3.1/24 to Spine2 Port2 10.0.3.2/24

 

So no need for STP since broadcasts won't cross into different networks = No switching loops?

 

And even more confusing to me is the Layer 2 implementation. If every Leaf Switch is connected with every Spine Switch, how are loops prevented? I've read about SPB / TRILL - Is this just used as better alternative to STP?

 

If so - Would that mean that for example Ubiquiti devices (EdgeSwitch) cannot be used for Spine-Leaf topology since (as far as I know) they do not support SBP / Trill on Layer 2 and also only have minimal L3 support so a L3 implementation is also off the table?

Thanks!

Link to comment
https://linustechtips.com/topic/1122298-spine-leaf-switch-topology/
Share on other sites

Link to post
Share on other sites

Spine and lead is a high level design for most data centers. Its main design purpose as you mentioned is to prevent loops but also for huge scalability and redundancy.

 

The basis around spine and leaf is likes are never directly connected and every node has a connection to every other node of the opposite flavor. You might think that this would lead to loops but actually if you look closer you never have a triangle loop. Since spines are not interconnected and neither are leafs, they cannot transit over each other which would indeed create a loop.

 

Quick sketch of what traffic looks like from one switch to another. As you see traffic cannot find a way to loop back. It can take multiple paths to reach the destination (red vs green) but its always a single hop.

image.png.9484f11db28e9747ebdbe058e69b757c.png

 

L2 - from L2 perspective, as described above, yes you will not need STP because there are no physical loops, every thing is just two hops away. Also look at the topology and how broadcast work, even if broadcast are sent, broadcast will never be resent out an originating port which as you see, there is only one uplink (port channel) further enhancing the loop prevention. For this you need mlag support throughout.

 

L3 - from a L3 standpoint, it varies. You can implement this with routing or switching, its not always one or the other. Most design now has moved to L3 all the things and is similar to what you mentioned in the OP. Now that BGP EVPN/VXLAN (L2 distributed over L3) is more common, its better to implement L3 everywhere for better load balancing and redundancy which L2 still has a hard time with. This allows you to utilize the multiple uplinks from the leafs to the spines to reduce congestion.

Link to post
Share on other sites

Thanks for your response!

 

So staying with L2, if my leaf is connected to three different spines all those ports would be a LAG on the leaf and MLAG on the spines? I guess this way it is indeed not possible to create a loop. 

 

As for L3 - I'm completely unfamilier with EVPN/VXLAN but I've already found a youtube series that should (according to someone else) be pretty good in explaining this.

 

If someone is interessted in the video:

 

Thanks!

 

 

 

Link to post
Share on other sites

4 minutes ago, instreamer said:

So staying with L2, if my leaf is connected to three different spines all those ports would be a LAG on the leaf and MLAG on the spines? I guess this way it is indeed not possible to create a loop. 

Its been a while since I looked into spine and leaf and I cant quite remember to specifics unfortunately. Based on theory alone I would say MLAG on spines.

 

12 minutes ago, instreamer said:

As for L3 - I'm completely unfamilier with EVPN/VXLAN but I've already found a youtube series that should (according to someone else) be pretty good in explaining this.

Take your time through VXLAN. There is a lot to it and make sure you are at least intermediate level to grasp a lot of its concepts. Its an amazing protocol that has a ton of uses even outside DC. 

 

Side note: Network Direction has some of the best network videos out there about explaining the theory around protocols. I highly recommend them .

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×