09-10-2016 03:47 PM
Within the fabric there are two load sharing methods for layer 2 forwarded traffic
1. Brocade trunk - this is proprietary and will only be present if the members of the trunk are connected to the same group. In this case traffic will be load shared on a packet by packet basis, which will result in a very efficient load sharing method. De-skewing at either end of the links prevent mis-ordered packets being received at the end station
On the 6740 VDX, the ports that are in the same groups and will form Brocade trunks are 1-16, 17-32, 33-40, 41-48, 49-50, 51-52
So if your ports between two VDXs are 1 and 8 then these will form one trunk. If you also have ports 17 and 20 then these will form a second trunk. The VDX will do per packet load sharing within each of these two trunks and it will do ECMP/flow based across the two groups resulting in a pseudo 40G connection between the two switches.
2. If the links are not in the same group then traffic will be load shared using ECMP. Load sharing will be flow based to ensure no mis-ordered packets. This won't be as efficient as Brocade trunk since a large flow could be on one link with no traffic on the other links. I realise this is an extreme example.
The port channel that you also mention is only used between the VDX and end station or to a switch outside of the VCS fabric. In this case load sharing is flow based. All members of the port channel could be on one VDX switch or, if the same port channel is configured on multiple VDX switches in the same fabric then you will have a multi-chassis lag which offers greater resilience. In Brocade terms this is called a vLAG. A differentiator of Brocade's vLAG is that you can have up to eight switches with members of the same vLAG, and you don't have to have a direct connection between those switches.