[j-nsp] Experience with two-tier MC-LAG?

[Karl Gerhard]

Hello

I would like to know whether anyone has deployed a two-tier MC-LAG:
MC-LAG #1 between the aggregation switches
MC-LAG #2 between the core routers
This would allow for a network design with "active/active everything".

Our current setup is the following: Access (Juniper EX), Aggregation (Juniper QFX), Core (Juniper MX).
Our access switches use Redundant Trunk Groups to connect to two aggregation switches.
Each aggregation switch has an AE where the first cable of the AE goes the MX #1 and the second cable of the AE goes to MX #2. To make this possible we have an MC-LAG between the MXs.
Pic: https://abload.de/img/single-mc-lag2io7w.jpg
With this setup only one uplink of the access switch is utilized, the other one is idle due to the nature of redundant trunk groups (active/passive).

If we had a second MC-LAG between the QFXs we could utilize both uplinks of the access switch.
Pic: https://abload.de/img/double-mc-lag4ppfd.jpg

Does anyone have experience with such a setup? Is this the highway to debugging hell or is it a good idea that increases bandwidth available to the access layer?

Regards
Karl