On the Bias in Active Measurements for the Estimation of Link Utilization