07-24-2014 02:48 PM
I have a doubt, we are designing a fabric which will consist in 4 "core fabric" switches and 24 "edge fabric switches" all L2....we will interconnect it to two MLXe-16.....Can someone tell me the difference between doing it with the MLXs with MCT and without? Why doing this? is there any advantage? Documentation says that you dont have to run STP without MCT while doing LAGs......
In the documentation says each MLX does a LAG to each VDX (two VDXs on the top of the fabric)....In our case we have 4 VDXs...this means we will have to do 2 LAGs per MLX towards the fabric....What is the best practice?
I cannot find the way to attach the images and the diagram but you can email me to firstname.lastname@example.org
08-06-2014 02:37 AM - edited 08-06-2014 02:43 AM
i can say so far: LACP on MLX/MCT is working fine if routing is on MLX. If routing is on VDX there is currently a issue.
The general advantage is if you use LACP you loadbalance and you don't need STP on VDX! I think thats very important because VDX with Trill and STP combination is very "new" .. and you normally be glad to have no STP on your vdx switches. We attached four vdx cores (two per dc) with one mlx at each DC and connected those mlx with MCT together. So you have a totally redundant design. If you won't use LACP you better use routing on vdx with routing transfer links instead of lacp. We currently have a L2 Link to VDX as Single link without redundancy because we have to enable STP on "core" vdx and we're a little bit carefully with that in production enviroment. Dunno if anyone have tested it some weeks to tell us how it works..