Same subnet across multiple routers

[emanuel32804]

Hi all,

I am trying to set up a network to have a ring topology for redundancy. Without it being a ring, the layout is simple; just switches with fiber links and it's just one big network. But once I close the ring, things obviously get more complicated. My goal is to have several VLANs with the same subnets present at all sites. At this point I figure having a router at each site and running some kind of loop-mitigation protocol is the way. I've settled on EIGRP, since one of the key things I want is load balancing around the ring. Attached is a picture of my topology. I only have one end user device in the diagram but imagine many being present on various VLANs. As far as the VLANs go, I'll just give some generic examples.

VLAN 10: 10.0.10.1/24

VLAN 20: 10.0.20.1/24

VLAN 30: 10.0.30.1/24

The idea is that these are the same subnets that can be accessed at all sites. It is especially worth noting that I would have a good amount of broadcast traffic going around too. I've been trying to wrap my head around this but I can't figure it out. It seems easy if I didn't care about the same subnets being present everywhere. Do I need to tunnel to every site?? Any thoughts?

Thanks!

[Kasun Bandara · Accepted solution]

three is basic thing that we cannot extend L2 sites over L3 networks without using any special protocols like VxLAN, OTV, etc. also you can use L2 VPNs to connect and extend same LAN to other site. you can use same VLAN IDs, but using same network IP ranges will not work. but make sure, you have designed network properly and understand why you need to extend same subnet within different sites. because, having different networks in different locations make it easier when you manages and control the connectivity between sites.

this is what i experienced. there may be different opinions from others according to their experience.

[emanuel32804]

Well that kind of confirms what I was thinking. The reason I need to maintain the same network across sites is because of the broadcast traffic. Would it be possible to do this using switching only? I know something like STP could block off the necessary ports to prevent loops, but then I loose the load balancing. And of course, STP is rather outdated as well. What do you think?

[MHM Cisco World]

Load balance' I see ring and only one link between sites' 

Is this lab or real network?

[emanuel32804]

Hypothetical for now but it will be a real network at some point. By load balance, I mean go either direction around the ring, i.e, don't just block off one link using STP and call it a day. I feel like it would be better to make traffic take the most direct route for two close sites as opposed to going all the way around the ring.

[Kasun Bandara]

@emanuel32804 what is the requirement for broadcast between sites? you may change it to multicast option, if available. yes if you configure STP, it will block redundant links to avoid loops. it will open closest link to send traffic (most cases). you can use something like VxLAN between sites. that also will do your requirement

[emanuel32804]

See below my response to Kanan, should answer your questions. But yes, it seems either multicast, VXLAN, or switches with STP are my options. Thank you!

[Kanan Huseynli · Accepted solution]

Hi,

the first my suggestion is if you have/ will have multiple sites with routers and routing between them, better to use OSPF than EIGRP.

Regarding VLAN extension, you should do over switching (legacy method) requires STP for L2 loop prevention. Or you should use normal method which is VXLAN, OTV which is also mentioned before. Just question, what is the reason for L2 extension between sites? If you have service that work on broadcast based, just check maybe it supports multicast too. Then you can safely run multicast routing over L3 environment.

[emanuel32804]

Out of curiosity, why do you recommend OSPF over EIGRP? I should mention all routers would be on one campus with PtP links as shown, just different areas within the campus.

I'll look into VXLAN and OTV, they might be what I'm looking for. The reason for L2 extension is that we use a number of protocols such as artnet that are unicast and broadcast only, not multicast. Some that we use such as sACN are multicast friendly though. The use-case is entertainment production where you have lighting, audio, video, etc going over the network. You do make a good point that if we can constrain to multicast only, then we can traverse subnets just fine. That will have to be part of our considerations.

Alternatively, like you said, switching with STP is the easy method. But it introduces latency and bandwidth concerns, so it will be the backup choice I suppose. Thanks for the help!

[Kanan Huseynli]

OSPF works better in ring topology than EIGRP (EIGRP Query process does not fit ring).

Whatever supports muticast, then use mcast for it. For others better unicast.

Also, if "site" in your network means segment of network where it has its own LAN/WAN etc. then you should definitely not use L2 between sites to avoid any L2 based issue.

By the way, VXLAN does not have strict clos (spine&leaf) topology. It requires IP reachibility between devices and control plane (RR, muticast even IngressReplication). If you have pure L3 between sites (between routers) with L2extension, then broadcast can not go over the links. Broadcast storm happens, then there is L2 ring without STP.

Support, siteA LAN sends broadcast, VTEP of siteA will send it to all others (with forwarding BUM traffic logic), other sites will send to downstream ports which should receive this traffic (based on L2VNI) and that's all. Storm can be inside LAN if STP is missed, not between sites.

In any case, better to have multicast, and unicast for unsupported systems.

[MHM Cisco World]

Vxlan is leaf and spine not rings and OTV with rings can not prevent broadcast storm.

In any case you need to BLK one link always to stop broadcast storm.

[emanuel32804]

Interesting. You make good points about the two, which I see after researching them some. I think that leaves me with either converting my traffic to multicast or using switches with STP.

[MHM Cisco World]

Yes'  suggestion otv or vxlan which can use to extended l2 over l3 but with rings any of them can not stop broadcast strom.

You need in end to enable stp and blk one link.

Run only multicast will prevent arp and mac learn between sites.

[MHM Cisco World]

to all that see this post 
the Loop occurs even between sites. 
MHM 

[Joseph W. Doherty]

I believe you have a few misconceptions about L2 and L3.

First, you cannot logically loop Ethernet.  I.e. a physical ring must be broken by something like STP or REP.

Second, there's no logical distance in L2.

Third, the shortest path LB you desire needs L3 and most dynamic routing protocols support it, although they differ in how they compute their metric.  The only feature unique to EIGRP is it can use unequal cost paths, proportionally, which wouldn't apply as described in OP.

One (ugly) approach might be to have all your subnets in one VLAN, but each subnet is per node and can only jump subnets via L3, which would choose shortest L3 path on a ring.  I.e. broadcasts would reach everywhere but L3 unicast wouldn't.  BTW, you don't peer all your (p2p) L3 (logically) on the same VLAN.