LCAP Etherchannel Throughput

[mattipler]

Good morning guys,

A very quick question. If I have an LCAP etherchannel consisting of 2 physical 1Gb/s interfaces. Does the channel get aggregated to 2Gb/s throughput? 

My understanding was that whilst theoretical throughput would be 2Gb/s, actual throughput would not achieve higher than 1Gb/s? Is this correct or is 2Gb/s achievable? 

Regards.

[Kanan Huseynli · Accepted solution]

Hi,

for the same "flow" which defined by port-channel load balance algorithm you will have maximum of 1 GB. Etherchannel cant provide load sharing (sharing traffic of the same flow between different physical interfaces). But in general you may have 2GB traffic in total as a result of different flow traffic.

HTH,

[mattipler]

Thank you. 

So from device A to device B, you would have a maximum of 1Gb/s over the channel. However, device C and device D may have a separate flow running concurrently at 1Gb/s which would bring the channel utilisation up to 2Gb/s. 

[Kanan Huseynli]

True, but if port-channel defines different physical links based on hashing algorithms. :)

Regards,

[Joseph W. Doherty]

As noted in other posts, a single "flow" would be limited the the capacity of a single link.

Between the two same hosts, usually all flows would map to the same link (because most hash algorithms use MACs or IPs) except if the hashing algorithm includes something like port numbers (which/if differ per flow). I recall (?) the 6500 series support port numbers as an Ethernet hash option. (I.e. if so, and if port number differ, you could obtain more than a single link's capacity between a pair of hosts.)

Even when the hash algorithm does well "randomize" Etherchannel link usage, two flows can still map to the same link. So in your example of hosts A and B and hosts C and D, each with a single flow between the two pairs, you might see the two flows on different links or the same link.

In the case of two links, in an Etherchannel bundle, with multiple hosts, rather than expecting twice the bandwidth throughput, you would expect about a 50% increase (again assuming "ideal" hashing).