— Dedicated infrastructures for end-to-end measurements are complex to deploy and manage. Equipment cost, the requirements for reporting bandwidth, and the administrative diversity of the Internet, are factors that potentially hamper scalability. This paper describes the architecture and implementation of an alternative approach in which the end-to-end probing and measurement reporting functions are embedded in a transport protocol, namely RTP. Suitably enabled hosts in a multicast group are effectively co-opted to form an impromptu measurement infrastructure. Coupled with our previous work on multicast-based inference, this enables the determination of the performance characteristics of internal network links of very large multicast distribution trees. Our experimental results show that an accuracy of about 1 part in 10 is attainable when inferring link loss rates in the 1% to 10% range, down to 1 part in 3 for loss rates down to 0.1%, this for probes generated by a regular audio so...
Ramón Cáceres, Nick G. Duffield, Tim