1

I have a bottleneck affecting traffic. There are 4 switches with each of them running a line to the next to provide internet. I haven't played around with LACP LAG too much and am wondering if the setup in this diagram would work. I am wondering if this would help my bottleneck between the switches.

enter image description here

4

The setup is totally valid, provided all devices support LACP LAG.

Two things though:

  1. The network design is not very good, neither for reliability nor from the performance/efficiency perspective. Basically, don't chain but build a tree. If possible, connect the top three switches to the bottom one directly (or even the firewall), shortening forwarding paths, removing dependencies, and removing the bottleneck from the interconnects.
  2. Note that aggregating links does not work like a single interface with the aggregated speed. Instead, each member switch uses an algorithm to select a physical link for each frame, so that similar frames always use the same link. Depending on the switch algorithm, the criteria can be source/destination MAC addresses, IP addresses, or IP addresses and TCP/UDP port numbers.

    So basically, the more flows run across a LAG trunk, the more likely it is that they distribute evenly across the physical links. Few flows are not likely to distribute well unless you fine tune the mentioned criteria.

If changing the topology isn't possible and the bottleneck at the bottom switch is hard to remove, using multiple VLANs with MSTP may be an alternative. Grouping the VLANs into several MSTP instances that you can then run over their dedicated interconnects allows you to guarantee bandwidth for the topmost switch which isn't (easily) possible with simple LAG trunks. If this is feasible depends on your workloads and their distribution across the switches.

  • If I were to use the bottom switch to bring in the internet from the firewall, could I distribute it to the other 3 switches using SFP to increase the bandwidth between them? – Paradox1911 Aug 15 '18 at 19:36
  • SFP as a fiber interface? The interface type increase bandwidth, it's all about the speed or bandwidth of the interfaces. If you worry about the distances between the switches then yes, fiber will cover much larger distances than twisted pair (10GBASE-SR runs over 400 m of OM4, or 10GBASE-LR over at least 10 km of OS1). – Zac67 Aug 15 '18 at 19:42
  • Basically if I bridged my firewall interfaces as one network then run a line from each switch to the firewall I would eliminate the chain and the bottleneck it is causing, correct? – Paradox1911 Aug 15 '18 at 20:01
  • Yes, that would give a star or tree topology with the firewall as the center or root of the tree. Two things to pay attention to: a) the FW might have to be configured as root brigde with the lowest priority of all bridges/switches involved b) switch-to-switch traffic would necessarily have to flow through the Firewall's bridging feature, which might turn out to be your next bottleneck. – Marc 'netztier' Luethi Aug 15 '18 at 20:40
  • We recently purchased a Watchguard T55 and I see the bridge feature to combine the various interfaces. Is this the root bridge you mean? – Paradox1911 Aug 15 '18 at 20:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.