Is it incorrect to have a layer-2 broadcast domain cross long distances or multiple sites?

  • Comments are not for extended discussion; this conversation has been moved to chat.
    – Ron Maupin
    Mar 16 at 19:45
  • @RonMaupin please post your answer from the comments. It would be helpful to the community. Mar 16 at 19:55

A VLAN is a virtualized concept of a link-layer segment. That segment is connected using physical-layer links. Those links can be directly extended to up to 200 km using commonly available parts, and far beyond that using special parts, or optronic or photonic amplifiers (esp. EDFA) - effectively, the limit is your budget or the size of the planet.

However, many protocols on top of L2 expect a low latency between 'neighbor' nodes within the same segment. They may fail to work as expected on geographically distributed segments.

A link of 1000 km has a minimum latency of 4.8 ms due to the speed of light - with an unlikely direct, optical connection, as the crow flies, using only photonic amplification. Usually, you'll get significantly higher latency due to more practical pathways, and electronics and buffering in the path. Employing a carrier (for MPLS or such) can also introduce serious fluctuations into the perceived link latency.

For instance, relying on xSTP for a redundant, loop-free topology may fail when the tree doesn't properly reconverge because of fluctuations in latency exceeding the design limits. Also, long links are expensive and redundant links blocked by STP don't earn any money.

L2 is often managed using broadcasting (e.g. ARP) which is highly inefficient on a large geographic scale. ARP might even fail or address conflicts may be masked because of high latency (when a near and a remote node share the same IP).

Also, with a partial failure/shutdown of one location (or possibly just a switch reboot) or its coming back online, all your L2 network is impacted.

In contrast, routed links can be utilized with much better efficiency. Using a proper routing protocol you can build a redundant mesh without the limitations of STP. ECMP and PBR enable better utilization of and better traffic control over your links.

And last but not least, any link-layer problem stays limited to the local segment and doesn't influence your entire network.

  • 3
    @RonnieRoyston A VLAN is a link-layer segment, the most common use for "segment".
    – Zac67
    Mar 16 at 19:38
  • Here's a Cisco documented deployment senario i just found for it, BTW Mar 16 at 19:54
  • That guide talks in no way about geographically separated sites. Of course, a VLAN configuration guide tells you how you can configure VLANs. It doesn't tell you how to design your network though.
    – Zac67
    Mar 16 at 19:57
  • it says In a typical deployment of VLAN mapping, you want service provider to provide a transparent switching infrastructure that includes customers’ switches at the remote location as a part of local site. This allows customers to use the same VLAN ID space and run Layer 2 control protocols seamlessly across the provider network. Mar 16 at 19:59
  • 1
    You can easily create a DMZ at each location where it's needed. Of course, it's a bit more complicated, but as I was trying to point out a routed network scales much better. Do as you like but you won't find many supporters for "one giant L2 network" here.
    – Zac67
    Mar 16 at 20:41

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.