Etherchannel use

he-man

Hey guys,

Just a quick question for those who work with production networks.

In a switched network with multiple redundant links between switches, due you create an etherchannel(s) with these links or just leave them as invidual links?

I can't see why you wouldn't utilize an etherchannel, because there combined use produces more bandwidth, a good thing right? or is there an underlying gremlin that could cause issues?

The reason i ask is that in alot of the material i've read, there doesn't tend to be much reference to etherchannels?

Cheers,

bighornsheep

It's for different purposes in redundancy, do you aggregate and use more to do more, or do you have literally redundant links for backups in case something happens.

If resource permits, you would probably have both. In my organization's network, all of our edge switches have multiple fibres to the core in two groups, one is for backup.

Netstudent

Ya like one primary bundle, and then a backup bundle. Each bundle could have any number of aggregated links supported by the hardware. Your redundancy protocols will treat each group or bundle as a single link, i think .

tech-airman

he-man wrote:

Hey guys,

Just a quick question for those who work with production networks.

In a switched network with multiple redundant links between switches, due you create an etherchannel(s) with these links or just leave them as invidual links?

I can't see why you wouldn't utilize an etherchannel, because there combined use produces more bandwidth, a good thing right? or is there an underlying gremlin that could cause issues?

The reason i ask is that in alot of the material i've read, there doesn't tend to be much reference to etherchannels?

Cheers,

he-man,

It depends.

Is there a SLA that requires redundancy over throughput?

joshgibson82

It also depends on your linecards....

If you have blade 1 and blade 2 that are not identical products, you can run into bugs with trying to bundle the links together. In particular, we saw issues using the GBIC port on a Sup1A and a 6408GBIC card. Even though they were the same speed in theory, we still had issues.

On the other hand, if hardware permits, etherchannel is definitely a positive thing to have because it doesn't require Spanning tree to reconverge upon a link failure.

Steve10393

joshgibson82 wrote:

On the other hand, if hardware permits, etherchannel is definitely a positive thing to have because it doesn't require Spanning tree to reconverge upon a link failure.

That's my thinking. I don't really understand what you're losing by having etherchannel on if you have spare links. Thing is, how do you know when one of those links goes down unless you go into the switch/router and check the interfaces frequently, especially on high speed links.

I don't know enough about the actual hardware on a switch/router, but would you also benefit from having more queues? Like if say you had 4 Gigiabit interfaces etherchanneled together, and it's overkill bandwidth wise, would you still benefit from it just because it's using 4 different queues hardware wise? Easier on the processor?

Netstudent

management protocols such as SNMP....Link status would proabbly be a simple trap. As far a queuing, you may be able to use some kind priority or customized queing for QoS that may give you some added granularity on traffic control. Either way, the processor will still compute the same amount of data unless your using CEF or somehting. But if the bundle is an overkill, then there may be no need to worry about queuing.