Inferring link-level performance from end-to-end multicast measurements
Rafaela Cáceres, Nick G. Duffield, Sue B. Moon, Don Towsley · 1999
We present MBone experiments that validate an end-to-end measurement technique we call MINC, for Multicast Inference of Network Characteristics. MINC exploits the performance correlation experienced by multicast receivers to infer loss rates and other attributes of internal links in a multicast tree. MINC has two important advantages in the Internet context: it does not rely on network collaboration and it scales to very large measurements. In previous work, we laid the foundation for MINC using rigorous statistical analysis and packet-level simulation. Here, we further validate MINC by comparing the loss rates on internal MBone tunnels as inferred using our technique and as measured using the mtrace tool. Inferred values closely matched directly measured values -- differences were usually well below 1%, never above 3%, while loss rates varied between 0 and 35%. 1 Introduction As the Internet grows in size and diversity, its internal performance becomes harder to measure. Any one orga...