Monday, August 3, 2020

VPLS.US 06/25/2011 using vpls for e-line

http://vpls.us/?p=756 Sat, 25 Jun 2011 01:48:59 +0000 timc
http://vpls.us/?p=756

using vpls for e-line

This is a pretty interesting question.

On the one hand you could use RFC 4447, using LDP to setup Pseudowire connections between two PEs. That’s pretty simple, straight forward, and with minimal amount of resources that get carved up by the PEs terminating the VC Virtual Connection.

Or, on the other hand, you could use 4762 to establish the point to points between the two CEs. Well there are trade offs for the least use of resources.

With RFC 4447:

  • You have to do some sort of migration if they want to go from ELINE to ELAN.
  • You don’t have any sort of MAC learning.
  • All BPDUs are automatically carried across the link.
  • It has the traditional carrier feel of a circuit being tied up.
  • No routing table is configured on the resources.
  • The turnup / build of a 4447 is MUCH simpler, in most cases the configuration is a single line of code.
  • The training / testing / operational aspects are MUCH simpler.

With RFC 4762:

  • No configuration Migration between ELine to ELan
  • Typically most Router Vendors will allow much deeper packet inspection of L2VPN instead of a Pseudowire. For instance I know of one larger router vendor that will allow classification via DSCP/TOS/Prec using 4762 but not with 4447
  • MAC learning can be turned off to make the behavior similar
  • Configuration is MUCH more complex
  • Traversing a Bridge-domain allows an additional touch point for troubleshooting exercises with the customer.
  • Adding additional sites is easy
  • Passing L2 protocols can require additional Configuration

There are trade offs for every scenario. I guess my main concern with using 4762 for building point to point ethernet is the situation where you are building EVPL or a Etree using point to points. Your trunk will have potentially hundreds of end points and building all of those VPLS 4762 tables seems to be a very large chore.

Then the other thing that has to be considered is the impact of the decision as it relates to your metro area. Turning off MAC learning at your PE doesn’t do anything to limit the amount of addresses that you could or will be seeing in your metro rings. Assuming of course that you don’t extend your pseudowires into the metro. That makes the decision of using VPLS in the metro even much more complex.

Personally, I’ve waffled between the different design theories, and have been swayed because of the features and capabilities of the vendors. However the end result to the customer is identical.

What is everybody else doing?
]]>