Dou Wen-hua

Design of a robust active queue management algorithm based on feedback compensation

Active Queue Management (AQM) is a very active research area in networking. The main objective of an AQM mechanism is to provide low delay and low loss service in best-effort service networks. In this paper we propose a new AQM algorithm based on the feedback compensation technique in control theory. The algorithm is called Proportional Integral based series compensation, and Position feedback compensation (PIP). By choosing the appropriate feedback compensation element and its parameters, the properties of the corrected system become dependent, to a great extent, on the series and feedback compensatory elements. Thus, PIP can eliminate the error incurred by the inaccuracy in the linear system model as well as eliminate the sensitivity to the changes in system parameters like load level, propagation delay, etc. The performance of PIP is compared to PI using ns simulations. From the experiments and analysis we conclude that PIP is more responsive to and robust under time-varying network conditions than PI.Active Queue Management (AQM) is a very active research area in networking. The main objective of an AQM mechanism is to provide low delay and low loss service in best-effort service networks. In this paper we propose a new AQM algorithm based on the feedback compensation technique in control theory. The algorithm is called Proportional Integral based series compensation, and Position feedback compensation (PIP). By choosing the appropriate feedback compensation element and its parameters, the properties of the corrected system become dependent, to a great extent, on the series and feedback compensatory elements. Thus, PIP can eliminate the error incurred by the inaccuracy in the linear system model as well as eliminate the sensitivity to the changes in system parameters like load level, propagation delay, etc. The performance of PIP is compared to PI using ns simulations. From the experiments and analysis we conclude that PIP is more responsive to and robust under time-varying network conditions than PI.

A Fault Tolerant Bufferless Optical Interconnection Network

In the high performance computer systems, optical interconnections are being used in recent years for improving the perform-ance. With the increase of the number of nodes in the system, as well as the increasing demand for high bandwidth and low latency, a few error nodes would result in the failures of whole interconnection system. So it is necessary for the optical interconnection to have the fault tolerance capability. A fault tolerant bufferless optical interconnection network named FT-BOIN is brought forward and the corresponding routing algo-rithms are brought forward. In the FT-BOIN, the intermediate nodes need neither to perform O-E/E-O conversion nor to buffer packets. This characteristic greatly improves the performance of the optical network. The simulation shows that FT-BOIN has outstanding fault tolerance performance, and can route packets in the network with nodes failure.

A hop-bounded single-actor selection algorithm for wireless sensor and actor networks

Wireless sensor and actor networks (WSANs) are composed of a large number of sensors and a small number of (mobile) resource-rich actors. Sensors gather information about the physical phenomenon, while actors take decisions and then perform appropriate actions upon the environment.In this paper, a single-actor selection algorithm for WSANs is addressed from hops and energy constraints. The corresponding optimal solution HBMECT (Hop-Bounded and Minimum Energy Cost Tree) is determined by integer linear programming (ILP). And then, a distributed approximate algorithm of HBMECT is provided.In the performance evaluation, the approximate algorithm of HBMECT is compared with MECT (Minimum Energy Cost Tree) and MPCT (Minimum Path Cost Tree) algorithms.Wireless sensor and actor networks (WSANs) are composed of a large number of sensors and a small number of (mobile) resource-rich actors. Sensors gather information about the physical phenomenon, while actors take decisions and then perform appropriate actions upon the environment.In this paper, a single-actor selection algorithm for WSANs is addressed from hops and energy constraints. The corresponding optimal solution HBMECT (Hop-Bounded and Minimum Energy Cost Tree) is determined by integer linear programming (ILP). And then, a distributed approximate algorithm of HBMECT is provided.In the performance evaluation, the approximate algorithm of HBMECT is compared with MECT (Minimum Energy Cost Tree) and MPCT (Minimum Path Cost Tree) algorithms.

TDOIN: A Novel Three-Dimensional Optical Interconnection Network for High Performance Computer

With the rapid development of high performance computers, the interconnection network performance has become a key factor restricting the overall performance. Due to the inherent limitations of the electrical interconnection network, optical interconnection has become a feasible way to improve the performance. In view of BOIN network, we proposed a Three-Dimensional Optical Interconnection Network (TDOIN), and introduced link control protocol and routing algorithms. Furthermore, experimental simulation based on the comparison of TDOIN and BOIN was carried out in case of latency and throughput. The results show that with the increase of network size, TDOIN network has more obvious advantages than BOIN network and can meet the needs of high performance computing systems.

Application of Residuation Theory in Network Calculus

This paper proposes a framework for the application of residutaion theory in network calculus. Min-plus convolution and deconvolution composes a dual-residuated pair, their properties can be studied by using residuation theory. End-to-end performance bounds can also be obtained by residuation method. The application of residuation theory in max-plus network calculus theory and time varying network calculus are also studied.

An efficient key agreement protocol in cluster-based MANETs

Key agreement in mobile ad hoc networks (MANETs) is a fundamental security service which provides essential protect to secure communication via open and shared media. Since MANETs are often made up of resource-restricted mobile nodes, the process of key agreement must be efficient, lightweight and minimize communication and energy consumption. This paper focuses on key agreement protocol in MANETs which is based on the Elliptic Curve Diffie-Hellman (ECDH) key agreement protocol and can prevent or resist various attacks. Due to the trust relationship among clusterheads, communication cost has been reduced eminently.

A Time Varying Network Calculus

Network calculus is widely used in network analysis and Quality of Service research, but till now the main work of the network calculus theory focuses on the time invariant field, and little work on the case that arrival curve and service curve are time varying functions. We propose a time varying network calculus theory, define the concept of time varying arrival curve and time varying service curve, obtain the time varying expression of delay bound, backlog bound and output burst. We also propose a method which use the closure operation to get the expression of a time invariant function, and use this method to analyze traffic clipper. The method proved its efficiency and simplicity.We also analyze the greedy shaper with non-empty initial buffer, and find that in this scenario, using time varying calculus is more simple.

Passive-Landmark Based Network Distance Prediction

This paper presents a novel passive-landmark based network distance prediction method—PLDP. Unlike other methods, landmarks in PLDP only need to respond to probes passively, while not measuring distances to other landmarks actively. Existing public servers, such as DNS servers, can be used as landmarks. So cost of deployment can be reduced greatly Experiment results show that both PLDP and GNP are efficient, having their own tiny advantages for different accuracy metrics. PLDP is more resilient to landmark failures, and much faster for prediction.

Modeling and Evaluating the Performance of the Bufferless Optical Interconnection Network (BOIN)

Most of the present optical interconnect networks within the high performance computers require either the buffering and opto-electonic conversion of the data packets or the pre-assigning of the optical path from the source to destination, which to a certain extent influence the performance metrics such as latency and throughput. Aiming at the limitation mentioned above, a hybrid optical-electrical interconnection for high performance computer system named Bufferless Optical Interconnection Network (BOIN) is brought forward. Basing on the topology and link control protocol of the BOIN, a performance model is setup. According to this model, the network throughput and delay are analyzed and evaluated. Simulation shows that BOIN has advantages over the other two networks. It can deliver high throughput at low latency, which can well satisfy the need for high performance computing systems.

A new systematic analysis and generation technique for the Internet topology

The dK-series analysis is an efficient technique to analyze properties of the Internet topology. Increasing values of d capture progressively more properties of the original topology at the cost of more complex states. In this paper, we present a new systematic analysis and generation technique based on the adjacency graph distribution, called dM-series analysis technique. The dM-series has less states and easier generation algorithm compared with dK-series, so it is more practical when analyzing large scale networks like the Internet AS-level topology. Experimental results show that the generated dM graphs converge to the original graph faster than the generated dK graphs when d increases.