Seeing the Unseen: Network Inference from Active and Passive Measurements on Wired/Wireless Networks
The Internet has grown into a tremendously large and complex system with an estimate of 300 millions end hosts and even more users. Many network paths involve multiple ISPs, with end hosts using access network technologies, such as Ethernet, Wireless LAN, cellular data networks, Cable modem, ADSL and dialup to access the Internet. An increasing portion of the Internet uses wireless, including wireless LAN and cellular data networks. Because of its size, complexity and heterogeneity, understanding and simulating the Internet has proved to be very difficult. Network measurements provide valuable insight and quantitative information on network performance, management, link and router provisioning and pathology diagnosis. My PhD research work aim to infer hidden network properties from the complex Internet by asking and answering the following five previously unaddressed questions based on network measurements:
· Is it possible to identify the existence of a dominant congested link (i.e., a significant source of performance degradation) on a end-to-end path based only on end-to-end packet delay and loss observations?
· Is it possible to classify an end user's access network using only end-to-end performance measurements?
· Is it possible to accurately simulate a network path based on end-to-end packet delays and losses observed on the network path?
· Is it possible to infer the extent of wireless LAN usage and identify wireless LAN traffic based on passive measurements at the edge of a large network?
· Is it possible to detect the use of wireless network performance enhancement approaches, such as Split-TCP, and quantify the performance gains based on end-to-end measurements in commercial cellular data networks?
These questions are important challenging issues in network management and engineering, network security, network modeling, network simulation and performance evaluation. My answer to all these questions is: Yes!. The following research projects answer the above questions.
· MINP: Model-based Inference of Network Properties
· ANTI: Access Network Type Inference
· TIMES: Trace-based Inference of Models for Emulation and Simulation
· WLAND: Wireless LAN Detection
· MICN: Measurement and Inference of Cellular Networks
|