2

We have bought two new switches that will be used as the core in our network. They will be directly connect with some LACP link. The aggregation layer consists of 5 L3 switches that will have one direct link to each of the core switches.

My question is how to best set up the vlans between core and aggregation. My first idea was to have one vlan for each link and have ptp connections between all agg-switches and the two core switches. But say I want some vlan to span the entire network I then have the problem of loops in the network if I want to use both links.

EDIT: Right now I have three scenarios in my head for the setup. And I want your opinion on the best way to setup a OSPF network with the structure shown in the diagram below.

  1. All router links are p2p. (/30 link nets)
  2. All router links are on the same subnet (and then using MC-LAG (DELL VLT) for redudancy and avoid loops)
  3. Stack the two core switches and use regular LAG for all links.

Network diagram: Network Diagram

5
  • I not sure about the question. But right now my thought is one vlan per aggregation switch and do L2 down to ToR.
    – Peter
    Jul 14, 2015 at 20:56
  • Let me rephrase, will you have hosts(vm's) that move around and need to maintain a fixed IP ?
    – Pieter
    Jul 14, 2015 at 22:46
  • Will your Tor switches dual connect to 2 x Agg switches ?
    – Pieter
    Jul 14, 2015 at 22:46
  • Im not sure any hosts will move around in the network like that. I think my "users" -> the non net-sysadmins want the convenience of having the same IP range all over the network. No, all ToR have one uplink to aggregation.
    – Peter
    Jul 15, 2015 at 8:23
  • Did any answer help you? if so, you should accept the answer so that the question doesn't keep popping up forever, looking for an answer. Alternatively, you could provide and accept your own answer.
    – Ron Maupin
    Aug 7, 2017 at 19:49

2 Answers 2

2

If your two core switches support vss, vpc, or mlag, you could simplify the network as the two core switches will look like one switch with a LACP ether-channel to the aggregation layer switches.

EDIT:

This will eliminate loops to the aggregation layer as each aggregation switch will have a single ether-channel link, while still allowing VLAN's to span all aggregation switches.

1
  • Yes, I know. I also found out the core switches are stackable. But right now I'm still opting for a L3 solution with p2p-links between every switch. But it would be very helpful to find some text on this subject.
    – Peter
    Jul 9, 2015 at 20:40
2

Don't do either of those. If you need to pass VLANs between the switches then they should be Layer 2 Trunks between the switches. If you need to exchange routing information you should connect P2P links between the switches and have them participate in your routing protocol.

Are all of your Aggregation switches connected in a mesh or are they in a series? What topology has you using 5 - 10 Aggregation switches? If you need this many switches you probably need a bigger switch. :)

In an idea scenario you would have 2 - 4 aggregation switches hosting the SVIs for your local data center. These numbers are fully dependent on the number of downstream switches that are required in each row. Then these aggregation switches would connect up to your core switches via Layer 3 Links.

You will run in to all sorts of issues if you try to implement what you are describing. Let us know more information about your scenario and we may be able to assist you further.

-Update-

The reason the switch models matter is because a detailed design will always be limited by the functionality of the devices that will be implemented. After doing some quick research on the Dell website I see that the S6000 switches are both stack and VLT capable. So here we have two paths to a more flexible and maintainable design.

If you stack the S6000 switches in the core they will become one switch with multiple modules. (Similar to a chassis switch, see this.) At this point you can configure links from both core switches to each aggregation switch in a non-blocking manner.

With VLT on the S6000 series you are able to configure an LACP Port-Channel from both of the core switches to each of the aggregation switches. They will appear to each aggregation switch as a single switch.

If your goal is to achieve Layer 3 at the "Aggregation" layer that you have described then the stack is your only option. You will not be able to create Layer 3 Port-Channel interfaces using VLT.

7
  • Well, we have clusters and they are all configured with a central switch, (this is the one I call aggregation switch. Then there are 5-7 ToR connected to that switch).
    – Peter
    Jul 2, 2015 at 19:59
  • One if the problems Im trying to solve is that we are going so have both private and public IPs in our data-center and it would be nice to have a single plan for the public ones (since they are few). But I have no problem splitting them up and route them if this is a better setup. Im not sure if I can describe then physical topology better than I did. We will have two core switches and want to have redundancy and full bandwidth between all clusters. So simply speaking I will have two 40G links from every cluster, one to each core switch.
    – Peter
    Jul 2, 2015 at 20:00
  • networkengineering.stackexchange.com/questions/19461/… talks about sort of the same thing.
    – Peter
    Jul 2, 2015 at 20:22
  • What kind of switches are these?
    – Daniel
    Jul 3, 2015 at 1:31
  • I don't see why switch brands would be relevant in a pure design question but right now the aggregation switches are HP5400 and one HP5900. the core switches will be dell s6000-on.
    – Peter
    Jul 3, 2015 at 6:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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