So you think you can measure IP QoS?

For broadband networking to be a commercial success, operators need to provide differentiated levels of quality of service (QoS). Networks are shared infrastructures of buffering and bandwidth resources, and are designed to bound the delays and losses inherent in packet queueing technology. But can customers be sure they are getting the QoS they are paying for? It is thought that active probing gives customers the ability to check that their service level agreements are being honoured, because it mimics "what the customer experiences". This may be an illusion. We investigate the trade-off between accuracy of QoS measurement and packet probing overhead. We show that, under realistic networking scenarios, measurement accuracy could be poor, giving little real indication of user experience. Our results demonstrate that the rate, size, and phasing of active probes must be carefully considered. Hence it is vital that measurement devices are configured appropriately for the networking and customer traffic scenarios.