[c-nsp] Basic Etherchannel Question

[Keegan Holley]

On Sat, Jan 15, 2011 at 10:33 AM, Phil Mayers <p.mayers at imperial.ac.uk>wrote:

> On 01/15/2011 12:42 AM, Peter Rathlev wrote:
>
>> On Fri, 2011-01-14 at 18:50 -0500, Keegan Holley wrote:
>>
>>> Just wondering what the general consensus was on hard coding vs.
>>> negotiating
>>> etherchannels.  I've always hard coded them and viewed the negotiation
>>> protocols as a possible point of failure.
>>>
>>
>> We always use LACP, since an unconditional port-channel connected to
>> something that's not a port-channel might lead to problems. I view it a
>> little like GE auto-negotiation -- I can't see a reason for not using
>> it.
>>
>
> At one time the Cisco "fast convergence" SRND recommended channel mode of
> "on" because it was a bit quicker bringing links up. We followed that
> advice, but TBH I've been reconsidering it lately.
>
> To an extent it depends on what *kind* of etherchannel you're talking
> about. If it's router->router and you control the fibre patching, a
> mis-patch is less likely.
>
> But if it's towards an edge server, where mis-patching gets more likely,
> LACP seems like a no-brainer.
>
>
I'm less concerned with connecting to the wrong device than with diagnosing
failures.  I've seen issues where a link in a hardcoded etherchannel stops
passing traffic but is not removed from the channel since there is no
negotiation protocol running.  Would dynamic protocols help here or is it
not worth the risk?  Just to be clear I'm talking about LACP, but I assume
PAGP is capable of the same.