MLS - Multi-layer switching

[Kevin Dorrell]

As risk of looking foolish, I confess I must have missed a trick with multi-layer switching, because there is something pretty fundamental that I don't understand.

Image a conventional campus architecture with a couple of switches in the distribution layer, lots of switches in the access layer, and a few VLANs. These VLANs are arranged organizationally, that is with each VLAN having a presence on several access switches, and each access switch serving n VLANs. (I prefer this architecture to the topological one with one per access switch because departmental traffic is kept in the same VLAN regardless of geography.)

As suggested, we do our inter VLAN routing in the distribution layer, either with a routing function built into the distribution switch, or with a router-on-a-stick.

So far so good.

Back in the olden days, these switches were all implemented with Cat 5000 series, so they could all do MLS. A distributed route-one, switch-many. We had a router-on-a-stick or an RSM. If a machine on VLAN x wanted to talk to a machine on VLAN y, then the first packet would go up to the router and back down again. But the MLS would observe this, and subsequent packets were switched by MLS rather than routed. As I understood it, this could happen even at the access layer. The first packet would go up to the distribution layer, up to the router, and back down to the access switch. Subsequent packets would take the shortcut, and be switched by the MLS. If the machines were on different access switches, this would happen in the distribution switch. If the two machines were on the same access switch, then it would happen in the access switch .

Now, re-implement this architecture in Cat 4500 series. Put the routing function in the distribution layer, and layer-2 switching in the access layer. Routing is switched off in the access layer, otherwise you end up with far too many routers to be manageable. Am I right in saying that all packets have to go up to the distribution layer and back again, even if they are destined for machines on the same access switch?

If so, have we taken a step backwards?

Kevin Dorrell

Luxembourg

[Prashanth Krishnappa · Accepted solution]

You are correct, with the implementation of 4500(or even 6500), all inter-VLAN packets would be routed at the distribution.

Only the 5500 series supports distributed MLS with supporting hardware. I would think this added value when 10/100Mbps was a norm and gig uplinks a luxury.

With the advent of gigabit/10 gigabit ethernet, CEF based hardware switching, I would not neccessarily think it is a step back.

[Kevin Dorrell]

Prashanth,

Thank you for the reply. That puts my mind at rest, 'cos I was beginning to think I had misunderstood something fundamental. I take your point about distributed MLS not being so relevant when you have Gig uplinks. That at least gives me something to tell my management if they ask the question.

Thanks again.

Kevin Dorrell

Luxembourg