Etherchannel on Cisco IOS Catalyst Switch

Hi Rene,

The way you explain network lessons are awsome but I do not find much info/lessons on Etherchannel.

It would be great if you can add a lesson that explains how LACP works in details with examples.

 

Thanks,

Rajiv

 

Hi Rajiv,

What exactly would you like to know about LACP?

Rene

Hi Rene…I don’t quite get how if we have the option of on or of in both lacp and pagp how the switches know which we mean…probably me being stupid. Tx

Hi Simon,

When you use the “on” method, we don’t do any PaGP or LACP negotation. It’s a “static” etherchannel configuration.

Rene

Hi Rene,

Trying to get my head around load-balancing.

In your example you suggest to change from (default) src-mac to dst-mac.

Does this mean that now the four computers go over one ether-channel channel? If so it will be a difficult act to get it right.

It might be easier/safer to upgrade to a 10G link as you suggested in your introduction.

Does the same principle apply for L3 load-balancing?

Also, if the four computers generate more traffic than one ether-channel channel can handle, will the ether-channel start to drop packages?

Hi Hans,

The default is src mac based load balancing.

From SW1’s perspective this is a great choice as each computer has a different MAC address. Traffic from the computers will be load balanced when it is sent from SW1 to SW2.

From SW2 to SW1, it’s another story. The MAC address of the router remains the same so using src mac based load balancing is not a good idea. All traffic will be sent using the same physical interface. It’s better to change it to dst mac address load balancing since the destination MAC address will be one of the four computers.

The same principle applies to L3 yes.

When your etherchannel is overloaded, you will see drops yes.

Rene

Hi Rene,

Thank you for the excellent explanation.

Re.:When your etherchannel is overloaded, you will see drops yes.

What I’m getting at is that the bandwidth you create with an etherchannel is not really the bandwidth you get in practices based on your load-balancing setting and type of traffic.

Best regards,

Hans de Roode

19 posts were merged into an existing topic: Etherchannel on Cisco IOS Catalyst Switch

Hi Hans,

That’s right. This sometimes surprises people…

For example, let’s say you have a fileserver and 100 clients. The fileserver is connected to the switch with 8x Gbit Etherchannel, the clients are using regular Gigabit interfaces. This will work well since there are many different client MAC/IP addresses, we can use this for load balancing and you might be able to get close to 8 Gbit of bandwidth for your fileserver.

We also have a backup server connected to the switch with 8x Gbit Etherchannel. During the night we copy everything from the fileserver to the backup server. What happens is that you will only get ~ 1 Gbit of bandwidth since the source/destination MAC+IP are always the same…Cisco doesn’t support round robin frame-by-frame load balancing for Etherchannel.

You will be better off upgrading the Gigabit Etherchannels to 10 Gigabit links on the fileserver and backup server in this case.

Long story short, building Etherchannels doesn’t really mean that 1+1 = 2 :slight_smile: It depends on the traffic…

Hope this helps!

Rene

Hi Rene,
You’ve confirmed my suspicion.
In stead of upgrading to a 10Gb link, would it not be easier to use routing, with EIGRP you have automatic load-balancing and L3 switches are more affordable. It also adds redundancy.

Best regards,
Hans de Roode.

Hi Hans,

For network devices that could be an option yes, L3 networks have some advantages over L2 networks…you don’t have to deal with spanning-tree, redundant links can be used instead of blocked and the convergence time is faster. For your end devices like a fileserver, you still have to deal with L2 interfaces (switchport).

Rene

Hi Rene

" Etherchannel will do load balancing among the different links that we have and it takes care of redundancy."

Load balance implies equal bandwidth distribution among the physical links. Would you agree that load sharing would be more appropriate, as equal bandwidth distribution is not always achieved due to the hashing algorithm?

Let me know. I appreciate your content, as it is always a reliable source for myself. Also, Nexus topics or ISE would be some a great addition if you are looking for new content.

-Mike

Hi Mike,

Load sharing might make more sense yes. If we could do per packet round robin then it would be more like load balancing since we could evenly share the load on all physical links.

With the hashing algorithms we can use for Etherchannel it can be hard to get above the bandwidth of a single physical link when there’s only a few sources/destinations. It’s similar to a highway, we can add more lanes but that doesn’t mean we can always drive faster :slight_smile:

Once i’m done with all R&S content I’m going to work on Nexus material, there’s quite some demand for it :slight_smile:

Rene

Goede Morgen Rene,

I re-created PAGP Etherchannel on 2 x 2960 X real lab based on example above but from your Vault Lab i…e blind. First I converted each of the 4 x physical interfaces to trunk mode then i configured using the interface range command two physical interfaces on sw1 to actively form the etherchannel :-

interface GigabitEthernet1/0/13
 switchport mode trunk
 channel-group 1 mode desirable
!
interface GigabitEthernet1/0/14
 switchport mode trunk
 channel-group 1 mode desirable

I then did the same on SW2 :

interface GigabitEthernet1/0/13
 switchport mode trunk
 channel-group 1 mode auto
!
interface GigabitEthernet1/0/14
 switchport mode trunk
 channel-group 1 mode auto

BUT the etherchannel did NOT come up so it stayed down until i did a shut and no shut on the P01 interface … can you tell me is this normal or what happened ? many thanks in advance

PRETTIGE DAG
Will

Hi Will,

This can happen yes. Sometimes it does come up, sometimes it doesn’t. Doing a quick shut/no shut on the PO interface is quick way to “refresh” all interfaces.

Rene

Would you use Etherchnnel between two L3 switches in the distribution layer (not utilizing/enabling VSS)?

Hi Jason,

If you can use VSS then yes, it’s a good idea. Only the higher end switches support it though.

Rene

Hi Rene,

You content is very rich dear:) I have one questions , If we use both end same like…

  1. Desirable--------Desirable
    2.Active--------------Active
    3.Passive------------Passive
    4.Auto----------------Auto

Is there any issue on technically ?? I use Active Ative and PO is running. Also Some Router Support LACP/PAgP.Could you please add the topic .Many Thanks

br/
zaman

Mohammad,
There are only a couple of combinations that will work if you set both ends as the same:

PAgP
Desirable - Desirable

LACP
Active - Active

Manual (not recommended)
On - On

Any other combination of having both sides set the same will fail to form an Etherchannel. This is because with both Auto and Passive, neither of those modes send out packets to create an Etherchannel. Instead they are only listening for inbound requests to create an Etherchannel. If you have two sides that are both listening, but not sending, no communication will actually happen.

Dear Andrew,

Many Thanks…

So If we set both end active-active or Desirable=Desirable then both side will ask for create Etherchannel to far end and no one will respond right and also that behavior will creat any problem???

I a have tried the Load Balancing topology with GNS3 by setting both end src-mac and see from my capture its taking all path from SW1 and SW2 when send pinging from MAC:AAA TO MAC:EEE and revert .Need more clarification on it.

br//
zaman