New yl/zl class w/o routed interfaces?

[André Beck]

Hi,

on a first look, the new 5400zl and 3500/6200yl seem to come closer to what I'd call a decent L3 switch for smaller routed core+distribution scenarios. They finally have multicast routing (was "forgotten" on the 3400cl) and actually can deal with multiple equal distant paths as one would expect (by load balancing them). More out of a sense of completeness I'd also checked the docs for routed interfaces, something the 5300xl class learned in later versions and of course expecting to find it too - but eh! The docs actually state the same thing they still state for the older routing switches: Nothing but VLAN switch virtual interfaces for routing.

This can't be true - or is it?

TIA,
Andre.

[Matt Hobbs]

I think you're right Andre, the new products behave the same as the earlier ones in respect to their routed interfaces.

What sort of interfaces would you like to see exactly? loopback interfaces? physical port based (traditional)?

I'm not that clued up on the advantages of one or the other to be honest. I have seen good cases for loopback interfaces but I've never really understood why I would want a physical based over a logical VLAN based routed inferface. (Especially since I can just configure 1 port to the VLAN).

If you have any scenarios that you could quickly share it would be appreciated.

[André Beck]

Hi,

> I think you're right Andre, the new
> products behave the same as the earlier
> ones in respect to their routed interfaces.

I was somehow brought to think the 5300xl got routed physical interfaces somewhere in 9.x or 10.x but that seems wrong, there is not a single trace of this so far in the release notes. I'll ask my colleague (who was at some technical presentation of HPs ProCurve crew in the valley) where he got this rumour. Maybe he (or me) mixed it up with 9000 series which have the feature due to their very different hardware base.

> What sort of interfaces would you like to
> see exactly? loopback interfaces? physical
> port based (traditional)?

I'm more out for routed physical IFs but having a loopback interface would just fit things well. In LAN design I can live without a loopback (though it is ugly, but quite often there isn't even IP address space planned for such things as loopbacks and transit networks, so why bother).

> I'm not that clued up on the advantages
> of one or the other to be honest. I have
> seen good cases for loopback interfaces

As a router is implicitely a multihomed device and any of its physical interfaces might be down at some point, having a common and never changing IP identity is very useful for a lot of things that happen in the control plane. OSPF router ID, management IP, source for syslog, sFlow etc pp - a lo0 would solve a lot of things.

> but I've never really understood why I
> would want a physical based over a logical
> VLAN based routed inferface.

If the box is running as a pure router, without any need (or even want) for L2 conglomerates, routed physical interfaces are simpler, more straightforward and less error prone. They especially tend to converge faster (e.g. link down/up doesn't have to propagate through a VLAN switch virtual interface to down/up a link state).

> (Especially since I can just configure 1
> port to the VLAN).

IMO this is a mess and an abuse of VLANs where they don't belong. What to do with the VIDs, reuse them on other L3 switches or not? How does the result look in PCM+? I simply feel a strong urge not to configure inter-VLAN routing when I really want transit networks between physical interfaces. Of course that would also benefit from real equal cost multipath (flow switched), which I don't think is available either (at least the xl docs explain in lengths that what they call ECMP is actually not per flow or per destination but per destination *network*).

> If you have any scenarios that you could
> quickly share it would be appreciated.

The basic "L3 switched core and distribution with L2 access" model (as to be found detailed in Ciscos SRNDs for fault tolerant campus designs) gives several good reasons why core switches should be L3 only and distribution switches should be L3 except towards the access layer. To be found on www.cisco.com/go/srnd and worth a look, IMO.

Thanks,
Andre.

[Matt Hobbs]

Excellent. Thanks for that. I wish you could give points to the person that actually asked the question in the first place.