Graduate Exam Abstract
May 5, 2008, 3:30p.m - 5:30p.m
Towards emulation of large-scale IP networks using end-to-end packet delay characteristics
Abstract: Network emulation combines concepts from network simulation and measurements and provides an emulated network testbeds over which application and protocol software can be evaluated. Network emulators allow the investigation of the interaction of network resources, protocols and applications in a controllable and repeatable manner. Existing network emulators are not scalable due to the limitations of available computer hardware infrastructure and the reliance on one-to-one packet mapping and modeling scheme. This research proposes a measurement-based modeling methodology for the design of a network-in-a-box emulator. The proposed methodology aims at overcoming the limitation of computational overhead in end-to-end network system characterization. A comprehensive study of end-to-end packet delay dynamics in the context of network system modeling is presented.
A framework for large scale IP network emulation, Overall Trend Replicating Network Emulator Tool (OTRENET), is presented. OTRENET intercepts data packet streams and modifies them in real-time based on network system models. The complexity and overhead of packet-by-packet mapping and modeling, while producing results consistent with measurements, is achieved by means of a traffic sampling algorithm. This algorithm monitors traffic metrics in a per-packet level to dynamically separate it into frames. Traffic behavior is then characterized by the average response of each time frame. The proposed Average Traffic Sampler by Time Frame Segmentation Algorithm captures the significant trends of the traffic metrics while not being sensitive to instantaneous fluctuations. Design, implementation and performance of the proposed algorithm and the emulator are described in detail. Experimental results are used to demonstrate the effectiveness of OTRENET in replicating realistic conditions imposed by modeled environments.
A detailed study of end-to-end packet delay dynamics is carried out in the context of network system modeling. Theoretical basis, techniques and measurements for characterization of network packet delay dynamics for different sending rates and network stages have been developed. A network system is modeled by means of characterization of the end-to-end packet delay and IPG dynamics taking into account the effect due to cross traffic, sending rate and packet size. Measurements of packet delays over the Internet under various conditions indicate that packet autocorrelation dynamics change according to the sending rate and packet size of the probes. Moreover, under weakly-stationary network conditions, traditional ARMA and ARIMA time series techniques can be used to model packet delay and IPG. Goodness-of-fit results under these conditions demonstrate the modeling accuracy for both packet delay and IPG processes for cases where the sending rate is relatively small compared to the link capacity. However, as the sending bit rate increases as a fraction of the bandwidth, IPG becomes a better alternative for network system modeling.
Measurement based analysis of packet streams has also demonstrated that packet autocorrelation, along with other packet delay characteristics, tends to vary in time in a non-stationary manner. A novel approach for online modeling end-to-end packet delay dynamics is proposed to address this. Proposed methodology models and captures the network system characteristics taking into account the non-stationarity of the packet delay samples, while keeping computational and storage requirements low. Experimental results demonstrate the potential for online packet delay classification with the proposed algorithm, while keeping computational and storage requirements low. The results presented show in general that analyzing packet delay processes by modeling the segmented traces yield a better understanding of the network system dynamics.
Adviser: Anura P. Jayasumana
Co-Adviser: Anura P. Jayasumana
Non-ECE Member: Yashwant K. Malaiya, CS Department
Member 3: Edwin K. Chong, ECE Department
Addional Members: Edwin K. Chong
Program of Study: