CES/CER can handle 1000 LSPs (lsp tunnels) before they stop making new ones. This can be an issue if you're letting your entire Cisco LDP domain pollute them. MLX would be a good order of magnitude larger but I don't have a specific figure.
Have tested and used MTUs of 9000+ for VLL with no issues.
Have tested and used MTUs of 4000+ for VPLS with some issues although this is the smaller CES platform and there may have been some feature gap in the Ironware version I was using at the time (MLX will have more features around this).
QinQ works fine if you do port mode VLLs. Also have tested it working in a "raw" VPLS environment (port mode again). In case that’s not clear, port mode would be where you have all client ports as untagged.
Raw mode every time if you ever want to do inter-op. This will set the VC-type to 5 and indicates that the port is transparent – from memory Cisco have this as a default in IOS.
I'm now trying to engineer it so we can add Juniper devices while changing as little Brocade config as possible. Can anyone help me with the following questions:
- How many LSPs can a Brocade MLX4/MLX8 handle before it falls over? We have a mesh of 15 chassis, and the output of "show mpls memory" looks like this:
Mem-Type Alloc BytesAlloc TotalAlloc TotalFree AllocPeak AllocFail FreeFail
LSP 28 25760 369 341 28 0 0
Does this mean that it's full? The documentation is sparse, I'm hoping someone else has gone down that rabbit hole and found an answer for me :)
- What MTU are people using for VLL/VPLS?
- Has anyone got QinQ working with VLL/VPLS on a Brocade?
- Does anyone have a preference for raw mode vs tagged mode for VLL and VPLS?