Summarization on distribution-to-distribution switches

[DaSt-1986]

Hi there,

I'm studying for my CCDP exam and came across an interesting discussion; if you create an campus network according to Cisco design guidelines with a layer three link between the distribution switches and a IGP, should you summarize the networks between these distribution switches.

Note; I'm talking about the link between the distribution switches, not between the Core and Distribution layer.

My arguments against summarizing these subnets, is that a local switch block usual doesn't contain much subnets, so summarizing wouldn't increase performance that much. Then why summarize?

I cannot find the correct answer in my CCDP book and the Cisco design documents don't provide me with an answer either. What do you think?

Daryl

[DaSt-1986]

Additional to this;

https://www.cisco.com/c/en/us/td/docs/solutions/Enterprise/Campus/HA_campus_DG/hacampusdg.html

First, it saids

Connect distribution nodes to facilitate summarization and L2 VLANs spanning multiple access layer switches where required.

Further on it saids:

It is important to summarize routing information as it leaves the distribution nodes towards the core for both EIGRP and OSPF.

[DaSt-1986]

Turns out this is a classical "just try it out"-situation. I build this in Packet Tracer to see what would happen. I build the following network:

I created a routed design between the distribution and summarized the LAN networks to 172.16.1.0/24. I advertise this summary via EIGRP to the Core and between the Distribution switches. The summary route is installed in the routing table of both distribution switches:

D       172.16.1.0/24 is a summary, 00:18:01, Null0

I use a looped layer two design between the distribution and access layer so I can use HSRP for the VLANs. I aligned the PVSTP root and HSRP priority for these VLANs so DIS01 is primary for VLAN10 and VLAN20 and DIS02 is primary for VLAN30 and VLAN40. This way, we do a little load balancing over the switches (although this results in assymetric routing from the core since it uses ECMP).

Now, when I disable a VLAN on one of the distribution switches, it removes the local connected subnet from the routing table, but a new route isn't added; the other distribution switch only advertises a summary route which doesn't get installed in the routing table since it already has a route with the same prefix to a local connected interface (Null0, as seen above). More specific routes are not advertised. This creates a black hole from the Core which is still using ECMP.

If I would use a layer three design between the Distribution and Access layer (which is a little more expensive but results in a much cleaner design), this problem wouldn't exists if I would use contiguous subnets on the access switches.

Conclusion: summarizing only makes sense at the edge of the layer three network going upwards, not side wards in the layer three network. Summarization at the distribution-to-distribution link can break things if the design is layer two so i would advice against doing it.

I attached the Packet Tracer file to the post so you can try for yourself (rename to a .pkt file)

[paul driver]

Hello

Would have thought stp inconjunction with the FHRP/timers etc would have flip over thus not losing connectivity especially the core utilising ecmp paths

Interesting read - cheers for sharing -

res

Paul