Posted on November 26, 2018 6:54 pm
 |  Asked by Parth Sakrikar
 |  63 views
0
0
Print Friendly, PDF & Email

hi Team,
please help to understand wrong speed aggregate issue not allowing to form the port channel link up in Arista switches.
what could be the reason for it ? and solution also.

0
Posted by Akshay Viswakumar
Answered on November 29, 2018 6:44 am

What ports are you trying to bundle? Are they running in different speeds?

0
Posted by Alexis Dacquay
Answered on November 29, 2018 9:15 am

Hi,

Can you illustrate your environment and explain why you would want to do so? (LAN, WAN, ...?)

It is generally not a best practice; how would you control the amount of traffic being distributed onto each link? A typical LAG shows that the bandwidth available is the sum of the physical members. In your case it would be false with 1Gbps + 10Gbps, it might seem as having 11Gbps available but it is not true.

Standard LAG load-balancing mechanisms involve hashing, so you might end up 50% on each link. That can exceed the capacity of the smallest link. If you have 4 Gbps of traffic, then you might have 2Gbps landing on the 1G link and 2Gbps on the 10G member. That would not work. There are extremely little cases when that would work; namely if the bandwidth is about just 1G all together. 2Ggps would not work because hashing is not 100% efficient, so you might have more traffic being distributed onto the link of smaller bandwidth.


If you are not going to use more than 1G bandwidth on the 10G link, why not put a 1G link in the first place?

What rule should be put in place for multi-speed: 100M with 1G, 1G with 25G, 10G with 100G, 1G with 400G? I think that from a design perspective, and for a neater operational support, having consistency is much more preferable.

So, since it does not work in almost all cases, it can be considered a mis-config or mis-design, and it is not allowed.

Also, is it not an IEEE standard that requires all the links in an LACP bundle have the same physical speed?


In summary, if you configure such mix speed LAG, the lowest speed link will not be participating in the LAG.

However, you might have very specific requirements, and if this is a MUST for some reason, then would you consider routing UCMP, Unequal Multi-path at Layer3?

https://eos.arista.com/eos-4-15-4f/bgp-ucmp/


In conclusion, the workarounds can be:

- change the links to have the same speed

- implement L3 UCMP

- do not aggregate (use active-passive with one of the 4 features: STP or port-backup or loop-protection or scripting(SDK orAPI))

- use BFD for fast failover (it might not be applicable to your use case, it depends on your design)



Best regards,

Alexis

Post your Answer

You must be logged in to post an answer.