The status is that we don't have an official Docker support in AMD, but we have been experimenting with it in the lab with no issues. Using native drivers, only (custom drivers could work, but not without tweaks in the code) - AMD works on Docker.
Of course you need to mind performance overhead added by Docker, it would be similar to virtualization, so not much (5-15%), but still some. Plus, running n AMDs in parallel will not sum up to the same capacity as a single AMD could achieve on the same hardware because of contingencies on the NIC buffers. So add another 10% overhead. But all of this may not mean much if traffic were low.
The bigger question is why run multiple AMDs for customers? A single AMD can segment traffic for consumption by different per-customer CASes, per. e.g. VLAN or software service. See https://www.dynatrace.com/support/doc/dcrum/rum-c... in DC RUM documentation. Once you have tenants defied on AMD, you can have different CASes set up to provide isolated reporting environments for customers. Would this architectural approach work? Its advantage is that there's less AMDs to manage.
OK, this sound reasonable. No custom drivers involved, and NetFlow collector doesn't use them. I guess there's a high likelihood of fully overlapping IP address spaces between customers (both client and server IP addresses), which means using separate NetFlow collectors per customer should be a sensible option.