An Evaluation Framework for Inter-Domain Link Congestion Detection Algorithms

Inter-domain link congestion can be caused by either under-provision or peer disputes. TSLP and M-Lab are two representative network-tomographic methods to detect congestion on inter-domain links. In this work, we propose a performancecost framework to evaluate and compare the two methods via a simulator built atop NS-3. Via detailed simulations, we obtained some insights on the two approaches and hope they can benefit further study in this area. Our experimental results have shown that TSLP is more responsive to link congestion and consumes less measurement bandwidth while M-Lab is capable of associating congestion direction to a congested link. Moreover, M-Lab may produce false positives due to its massive measurement traffic and underpinned TCP congestion control. Our simulator source code is also available online for the research community.