Fig. 1. Overflows / minute over the night of 7/30/01. The overflow rate was high (usually it’s zero) because files were being compressed in the background.

Fig. 2. Threshold over the same period.

Fig. 3. Event rate over the same period. The target is 60. The actual average is much higher because the threshold can’t respond fast enough in noisy conditions – I’ll try a larger step size.

Fig. 4. Response functions for different numbers of samples