Layer 3 redundancy; individual layer 3 links vs Portchannel

[gschmitt.ngit]

Does anyone have evidence as to which technique is "better" (more reliable, faster)?

Cat6500, dual SUP720, native IOS mode.

Partial topology:

Sw_1 (5/1)----Layer3 link----(5/1) Sw_2

Sw_1 (5/2)----Layer3 link----(5/2) Sw_2

In the above scenario, there will be 2 equal cost layer3 path from Sw_1 to Sw_2. So, on any link failure, we rely on IGP protocol (in our case, OSPF) for redundancy.

If we have something like,

Sw_1 (5/1)---L3 port-ch----(5/1) Sw_2

Sw_1 (5/2)---L3 port-ch----(5/2) Sw_2

Then, we have to rely on channel hashing algorithm for redundancy.

[Aaron Harrison]

Hi

I did something like this this week - we found that both worked well, although the failover between the links wasn't great with port-channels.

We have two 100Mb Ethernet presented circuits between two sites - we configured as an L3 port-channel (using 3750s at either end in this case). If you hard code them as on (channel-group 1 mode on) and the link went down beyond the NTEs the ethernet link to the switch never dropped, so the Etherchannel never failed the unavailable link.

If we used PAgP/LACP the switches either end did detect the failure, however it took 2 minutes (presumably 4x the hello time) - this didn't seem to be tunable (at least on that platform).

We ended up using OSPF to control the links - which failed over in a few seconds and also performs even-cost load balancing.

With regard to load balancing FEC bases it's choice of path on source MAC, dest IP or some other factor - although this is configurable on some platforms you need to think it through and trial it for your traffic platforms to get near-even load balancing.

Same applies to routing protocols really - you can per-packet load balance which should be pretty even but it's best avoided if using latency sensitive stuff like voice...

Regards

Aaron

p.s. please rate helpful posts :-)