MLAG, LACP & ISC Blocking Filters

[nixcraft]

I'm learning about MLAG and would like to confirm my understanding. Here's a diagram for reference.

The load sharing in steady state for traffic coming from upstream and going to the server is completely determined by the upstream. Meaning that unicast traffic coming from upstream links will reach the server on that sides connection only.

Now if one of the upstream connections were to fail, traffic coming from upstream would still only reach the server on the side still operating and would not cross the ISC links due to blocking filters.

So the only way to avoid the bottleneck in that scenario would be to either add new cross-links from the upstream or additional links to the server's MLAG member ports.

Having the switches in a stack would avoid this bottleneck as the stack link would be utilized, but that brings other considerations regarding stacking in general.

Am I understanding this correctly?

[OscarK]

I think you understand right, if the left 25 Gb uplink would go down, we would only receive traffic on the right 25gb link and that would all be switched to the direct attached 10gb link. If there is more than 10gb of traffic that would be dropped, we will not use the ISC for that.

[nixcraft]

Thanks Oscar. By the way, I read somewhere someone saying that single-attached devices aren't recommended to be connected to MLAG peers. Why would that be? Perhaps it only meant that it isn't best practice as those devices might possibly use the ISC links more heavily.