LAG Questions

  • 0
  • 1
  • Question
  • Updated 2 years ago
  • Answered
Hi!

I want create a LAG configuration, but i have some questions.

1. can i configure it using ports to diferent slots?
2. once configured if the master port pass to down state , everything LAG link is down?

Regards
Photo of Daniel Valera

Daniel Valera

  • 734 Points 500 badge 2x thumb

Posted 2 years ago

  • 0
  • 1
Photo of Patrick Voss

Patrick Voss, Alum

  • 11,574 Points 10k badge 2x thumb
Hello Daniel,

You can use any of the ports in a stack or a chassis to create a lag. If the master port goes down a new master port will take over.
Photo of Nick Yakimenko

Nick Yakimenko

  • 2,416 Points 2k badge 2x thumb
1) within one stack - yes
2) no, you create virtual port - LAG - and it is "link up" while one of the ports, that belong to LAG, is linked
Photo of Ijaz Chughtai

Ijaz Chughtai, Employee

  • 220 Points 100 badge 2x thumb
Hi Daniel, 
Yes, you can have ports from different slots as long as the ports are off the same speed ... you cannot mix 1G with 10G. 
If the master port goes down, one of the member ports takes over the LAG mastership. So having one port go down in a LAG does not break the entire LAG.

Hope this helps .
Photo of Carsten Buchenau

Carsten Buchenau

  • 942 Points 500 badge 2x thumb
Hi Daniel,

I assume this is about EXOS firmware (Summit switches or BD chassis). Yes, of course you can build a lag with ports on different blades / stack members. They have to be the same speed though, e.g. 2 x SFP 1 Gig. And when the Master goes down, the LAG remains active!

For the record, with EOS you might want to enable singleportlag, to ensure that a 2-port LAG still remains active when 1 link fails / gets removed.
Photo of Erik Auerswald

Erik Auerswald, Embassador

  • 12,886 Points 10k badge 2x thumb
EOS should be configured to disable LACP on all ports that are not intended to be used as part of a LAG. This is the default on the current hardware.

The "singleportlag" feature is needed after a LAG goes down completely (e.g. switch reboot) and then only one link comes up. Without "singleportlag" enabled, the link will not form a LAG, with "singleportlag" enabled it will.

With the older default of LACP active on all ports, single links connecting switches would form a LAG, until the number of supported LAGs on the switch was exhausted. The LAG number (lag.0.X) would be random, the configuration would often not be correct. Therefore EOS requires two links to form a LAG by default. It will still keep an established LAG as long as at least one participating port is active.
Photo of Daniel Valera

Daniel Valera

  • 734 Points 500 badge 2x thumb
thanks guys!

so i can with a simple configuration? for example

SW1: enable sharing 5:1 grouping 5:2 6:1-2 algorithm address-based lacp

SW2:enable sharing 5:1 grouping 5:2 6:1-2 algorithm address-based lacp
Photo of Patrick Voss

Patrick Voss, Alum

  • 11,574 Points 10k badge 2x thumb
The grouping needs to include the master port:

enable sharing 5:1 grouping 5:1-2,6:1-2 lacp

Unless you want to specify a different algorithm it will default to L2
Photo of Daniel Valera

Daniel Valera

  • 734 Points 500 badge 2x thumb
thanks for the info Patrick
Photo of Carsten Buchenau

Carsten Buchenau

  • 942 Points 500 badge 2x thumb
And in most cases you want to set the algorithm to L3_L4
Photo of Mel78, CISSP, ECE

Mel78, CISSP, ECE

  • 1,044 Points 1k badge 2x thumb
MLAG even more flexibility
Photo of Daniel Valera

Daniel Valera

  • 734 Points 500 badge 2x thumb
Last question: I need to restart computers to take the LAG configuration?
Photo of Patrick Voss

Patrick Voss, Alum

  • 11,574 Points 10k badge 2x thumb
Hello Daniel,

I just need some clarification. By computers are you referring to the switch? or are you referring to a server on the other side. You should not need to reboot either for the LAG to take effect. LACP does need to be configured on both sides before the ports become part of the LAG.
Photo of Daniel Valera

Daniel Valera

  • 734 Points 500 badge 2x thumb
Yes Patrick i ́m referring about Switch... tanks a lot.

regards
Photo of Daniel Flouret

Daniel Flouret, Employee

  • 7,470 Points 5k badge 2x thumb
Daniel,

Starting with EXOS v15.7 you have the possibility of
configuring the minimum number of links that must be active in a LAG before it is considered as "down".

Suppose you have two LAGs, each with 4 links, and that you configure one of them to be a backup of the other with the
configure ports primaryPort redundant secondaryPort {link [on | off]}
command.

The
secondary port (LAG) will not be activated until the first port (LAG)
goes down. In a normal LAG that would happen only when ALL  the links in
the primary LAG go down. But if you have, say, 4 links in each LAG, you
may want to switch to the secondary port (LAG) when two links in the
primary LAG fail, so you don't lose performance.

You would do that with the
configure sharing port minimum-active min_links_active
command.

Also, in EXOS v21.1 (and soon in EXOS v16.2) there's a new feature specifically for when you have ports on different modules in a modular switch, or different switches in a stack, that lets you modify how EXOS distributes traffic among the members of a LAG.

in the past, EXOS would always distribute traffic to all active members of a LAG. Now, you can specify a subset of the active member ports as elegible on a per slot basis. using the
configure sharing distribution-mode
command.

You can specify the "local-slot" keyword to restrict distibution of traffic only to ports that are in the same slot (switch) where the traffic has been received. This prevents forcing traffic to traversethe backplane (or stack port)s to be transmitted through a LAG port in another slot (switch) in the chassis (stack), reducing latency.

You can also create "port lists" to indicate a subset of ports through which the traffic would be transmitted.

This only applies to unicast traffic.