06-18-2014 05:49 AM - edited 03-10-2019 12:27 PM
Hello,
Wikipedia (http://en.wikipedia.org/wiki/100_Gigabit_Ethernet) states that 40 GE transceivers use 4 optical lanes for 40GBASE-SR4 and 4 wavelengths for 40GBASE-LR4, each with 10.3125 Gbit/s. 100 GE transceivers work in a similar way.
Does anybody know how this effects tcp/udp flows; can a single flow use 40GE or only 10GE in this case?
How is the traffic distributed on the 4 lanes/wavelengths, similar to port-channel hash method?
Where can I find some documentation in the web concerning this?
Many thanks in advance
Thorsten Steffen
06-18-2014 08:21 AM
I was able to find one reference in a whitepaper specific to the Cat 6k.
It states:
"Independent flows are load-balanced across the four ingress channels, using a configurable hash algorithm similar to EtherChannel. The default hash input is src-dst-ip.
Note: The 40 Gigabit Ethernet input hash computation is independent of (and/or in addition to) the EtherChannel hash computation of the Forwarding Engine ASIC (PFC4/DFC4)."
I would venture a guess that similar setup applies to the other platforms.
06-22-2014 11:24 PM
I already read this document before. In my understanding it states the method of sharing ingress traffic of the 40GE module to the four internal 16GE lines.
Maybe egress behaviour is similar, unfortunately it seems that there is no documentation in cco concerning this.
Discover and save your favorite ideas. Come back to expert answers, step-by-step guides, recent topics, and more.
New here? Get started with these tips. How to use Community New member guide