A) Reduce the number of miners per controller.
For best performance, no more than 2 miners per AUC chain. It will drastically lower your number of Stales.
A couple reasons for that: the main one is that the RasPi-1, 2, and 3 LAN circuit goes through the USB com channels. BTW, RasPi-4 fixed that and uses a dedicated connection between the network interface and the CPU - part of the reason they have better performance. The other reason is that is takes a fair bit of time (several 10's of ms per-miner on the AUC) to send/receive data between a miner and controller. Having more than 2 on a chain slows things down a lot. The baud rate of each AUC is 400,000 B/sec. That is a large cause of Stales seen by a pool. A cluster produces significantly more stales as opposed to what say an A10/A11/A12 produces or even a 921 by itself.
B) Use a program such as
Awesomeminer to monitor the miners.
Along with looking at each cluster on a controller as 1 big miner, AM has a toggle specifically for handling how the Avalon 7xx, 8xx, and 9xx work in clusters to let you look at each machine by itself. I highly recommend it. Their support thread in the Forum is
https://bitcointalk.org/index.php?topic=676942.0