One way to work around it is to set "shares" to a value (let's say 1000000, depending on diff on your pool), then run your mining in "while" loop - for example -
Code:
while true; do sgminer -c $HOME/.sgminer/sgminer.conf ; sleep 5 ; done
That way even if some of the modules drop out, the miner will restart after reaching preset number of shares with remaining threads anyway and reinitialise all modules after 5 second rest.
Another alternative would be to run each board as separate mining thread per device serial in 'screen', see which ones are flakey, downclock them for better stability etc
