Research On Lightweight Distributed Topology Measurement Technology For Local Networks

With the complication of network structure,more and more attention has been paid to the related research of network topology measurement.At present,the related research work of network topology measurement mainly focuses on: improving the underlying topology measurement technology,laying out high-availability measurement points to build distributed measurement platform,and improving the ability to analyze topological data.Internet consists of multiple local networks(states,AS,etc.).Therefore,we can analyze the topology of the local network before analyzing the Internet topology.Currently,there are numerous free-usage Looking Glass servers in the world that provide traceroute measurements.These servers cover a wide area and can be used as measurement points to building the topology measurement platform.To improve the efficiency of topology measurement for local networks,we design lightweight topological network topology measurement tasks,including the destination IP address should be representative in local network,and rational scheduling of measurement tasks should be optimized and prediction of topological measurement results.Based on above-mentioned proposal,this dissertation focuses on the following aspects of the topology measurement in local network:(1)Building a lightweight distributed measurement platform,we firstly collect a large number of Looking Glass servers and observe the structure of measurement point interface to design a set of rules.Then,since measurement point automatically generates configuration file by rules,we reasonably dispatch high measurement points to concurrently perform measurement tasks.At last,we provide visualization of some of the measurement results,The topology discovery capability of this paper has been improved by nearly double.(2)To meet the needs of lightweight measurement,we select the destination IP address set of the local network based on the historical topology data.First,analyzing the topology features of the local network by the historical topology data.Then,we divide the IP address of the local network into multiple IP clusters,and select the representative node to join the destination IP address from the IP cluster.At last,by compare with the traditional algorithms that generate destination IP address sets to discover topology discovery capabilities of our algorithms.(3)In order to use a small number of measurement tasks to discover a large number of IP addresses of local network that connected to external network,we invent a method to select tasks based on semi-supervised clustering.First of all,we analyze the topological characteristics of local network's IP addresses that connected to external network.And then,we propose a semi-supervised clustering algorithm based on measurement task's IP address attributes and part of measurement data.At last,we select tasks from samples that cannot be clustered to perform,until the IP addresses of local network that connected to external network is no longer found.We can find 90% of local network's IP addresses that connected to external network by 3% of the measurement taskFor the research methods mentioned in this paper,we have designed experiments to analyze its topology discovery capabilities.Through measurement experiments,it is found that the measurement platform we make has a good ability of topology discovery.Moreover,the destination IP address selection algorithm and measurement task selection algorithm greatly improve the measurement efficiency.