IP addresses distribution in Internet and its application on reduction methods for IP alias resolution

Abstract

Discovery of Internet topology is an important and open task. It is difficulted by the high number of networks and internetworking equipments, and even by the dynamic of those interconnections. Mapping Internet at router-level needs to identify IP addresses that belong to the same router. This is called IP address alias resolution and classical methods in the state of the art like Ally need to test IP addresses in pairs. This means a very high cost in traffic generated and time consumption, specially with an increasing topology size. Some methods have been proposed to reduce the number of pairs of IP addresses to compare based on the TTL or IP identifier fields from the IP header. However both need extra traffic and they have problems with the probing distribution between several probing nodes. This paper proposes to use the peculiar distribution of IP addresses in Internet Autonomous Systems in order to reduce the number of IP addresses to compare. The difference between pairs of IP addresses is used to know a priori if they are candidates to be alias with certain probability. Performance evaluation has been made using Planetlab and Etomic measurement platforms. The paper justifies the reduction method, obtaining high reduction ratios without injecting extra traffic in the network and with the possibility to distribute the process for alias resolution.

On the performance and improvement of alias resolution methods for Internet core networks

“…The method is based on the characteristic distribution of IP addressing in Internet routers presented in [28]. IPoffset parameter is defined as the absolute value of subtracting one IP address from another |IP1 − IP2|, considered both as 32-bits unsigned integers.…”

“…Due to the Internet organization in AS and the addressing allocated to each AS, IP addresses that belong to the same router have concrete IP-offset ranges that can be used to predict which IP addresses have more probability to be aliases. IP-offsets with more probability to be aliases are centered around 0 and 2.14 * 10 9 (a half of IP addresses range 2 32 ) as stated in [28]. A clustering algorithm is used to find those and other IP-offset intervals that imply significant probability of aliasing.…”

“…This means that IP-offset ranges with more probability to be aliases can be pre-calculated in certain networks, and those intervals are valid in another totally different network. In [28], the training network has close to 100 thousand pairs, and clusters generated are perfectly applicable to another network with close to 1.5 billion pairs. Calculating the clustering in the same network where aliasing is going to be performed only provides a marginal improvement.…”

“…The IP addressing organization provides a specific behavior in this sense, and it is independent of the network under study [28]. In our case, we use Etomic to obtain the cluster definition.…”

“…In a real topology discovering process, routers in access networks and different core networks should be traversed, providing more variability in the IP addressing schemes used and Fig. 12 IP-offset reduction method results using Etomic clustering over access+core networks therefore making useful this reduction method [28]. Using 50 Planetlab nodes as probing stations, traceroutes were made between all of them.…”