Measurement based inter-domain traffic engineering

(draft slides are available here: https://www.dropbox.com/s/r7czxo7jefk147o/NANOG67__Measurement_based_Inter_domain_TE.pdf?dl=0) This presentation concentrates on …

(draft slides are available here: https://www.dropbox.com/s/r7czxo7jefk147o/NANOG67__Measurement_based_Inter_domain_TE.pdf?dl=0) This presentation concentrates on an inter-domain traffic engineering scenario for multi-homed stub ASes. It involves making best use of the multiple egress transits that are available. The idea is to dynamically choose the best egress transit for each potential destination based on network measurements. We highlight the benefit of such scenario by showing the performance difference of different transit providers. The building blocks and challenges in realising such a system is revealed. Especially, how to deal with that huge amount of destination BGP prefixes is of crucial importance to the scalability of such design. We aim at sharing our observations based on working traffic traces from networks of diverse profiles (e.g. ISP, content provider, etc.) located in different countries (USA and some other European countries.) Our results show that we can predict prefixes that cover a satisfying amount of traffic with much simpler methods than those mentioned in previous studies. Further, intelligently making route decision requires proper interpretation of network measurements, especially those on RTT. First, we discovered that not all RTT measurements are eligible for inter-domain traffic engineering. We propose to briefly demonstrate our approaches in revealing the origins of “extra” variations in RTT measurements. Second, we demystify one common doubt on RTT measurements: do AS path changes cause major RTT variations? We show that it is not necessary the case with recent measurement data from RIPE Atlas.