• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • Public Review for Estimating Network Proximity and Latency Puneet Sharma, Zhichen Xu, Sujata Banerjee and Sung-Ju Lee Given all the previous work, is there space for improvement in this problem? This paper shows that this is actually the case. First, Sharma et al. focus on a different version of the problem: what if a client wants to find out the nearest server that offers a certain service without knowing the actual IP addresses of the can- didate servers? Second, even though their scheme, called Netvigator, requires some landmarks, they find a practical way to increase the set of nodes that are used as landmarks based on latency responses from the routers along the path to each landmark (the responses can be collected with a tool such as traceroute). This simple idea can have a major impact on accuracy, without requiring additional installed infrastructure in the form of more landmarks. The paper also presents interesting measurement results from two actual deploy- ments, one over Planetlab and another over a large corporate intranet. The reviewers were mostly positive about the paper and they liked the approach of using traceroute and router responses for expanding the latency information provided to the clustering algorithms. They also expressed some concerns, however, that the reader should be aware of. First, the comparison of Netvigator with GNP and Vivaldi may be unfair, to a certain degree, for the reason that Netvigator uses a larger num- ber of effective landmarks (when we consider the router-milestones as landmarks). Similarly, the overhead comparisons are also not obvious, given that Nevigator introduces an indirect overhead due to router ICMP responses. Finally, the reviewers also pointed out that a comparison with Meridian would be more appro- priate, as both Netvigator and Meridian take the approach of avoiding coordinate systems in order to find the closest server to a given node. As it usually happens with problems that have already seen several possible solutions, a quantitative com- parison with previous work is important, but also quite difficult. It is possible, in my opinion, that the solu- tions that eventually prevail in the field are not always the most accurate or elegant but those that satisfy the needs of actual applications in a practical way. Public review written by Constantine Dovrolis Georgia Institute of Technology a c m s i g c o m m ACM SIGCOMM Computer Communication Review 39 Volume 36, Number 3, July 2006