Seyed H. Hosseini

Analysis of a graph coloring based distributed load balancing algorithm

Abstract We present the analysis of a distributed load balancing algorithm based on edge coloring of undirected graphs. One version (linear version) can be studied directly using linear system theory. We show that the performance of another version (integer version), which is more realistic in that the loads are integers, can be studied as a perturbation of the linear version. Both versions of our algorithm converge to stable behavior for arbitrary topologies. In the case of the binary n -cube processor network we prove that after n steps of the integer version, for any initial load distribution, each processor has a load not more than n /2 away from the average.

A stochastic model for beaconless IEEE 802.15.4 MAC operation

IEEE 802.15.4 is a popular choice for MAC/PHY protocols in low power and low data rate wireless sensor networks. In this paper, we develop a stochastic model for the beaconless operation of IEEE 802.15.4 MAC protocol. Given the number of nodes competing for channel access and their packet generation rates, the model predicts the packet loss probability and the packet latency. We compared the model predictions with NS2 simulation results and found an excellent match between the two for a wide range of the packet generation rates and the number of competing nodes in the network.

On fault-tolerant structure, distributed fault-diagnosis, reconfiguration, and recovery of the array processors

A study is made of the design of fault-tolerant array processors. It is shown how hardware redundancy can be used in the existing structures in order to make them capable of withstanding the failure of some of the array links and processors. Distributed fault-tolerance schemes are introduced for the diagnosis of the faulty elements, reconfiguration, and recovery of the array. Fault tolerance is maintained by the cooperation of processors in a decentralized form of control without the participation of any type of hardcore or fault-free central controller such as a host computer. Time redundancy is utilized by assigning the functions of the failed processors to fault-free processors. >

Distributed Fault-Tolerance of Tree Structures

Tree structures, as the interconnection structure in networks of many processing elements, have interesting features such as regularity ease of expansion, simple routing, simple addressing, suitability for VLSI/WSI implementation, etc. Distributed fault tolerance of these networks is considered. It is assumed that in these structures, there does not exist any central failure-free entity for providing services such as diagnosis of faulty components, system reconfiguration after failure, control, or coordination among the processing elements. Every processing element is able to diagnose the condition of every other node or internode communication paths via a truly distributed scheme.

On self-fault diagnosis of the distributed systems

The problem of achieving fault diagnosis in a network of interconnected processing elements (called nodes) is considered. It is assumes that there is no central facility to control, coordinate or mediate among the processing elements. Every node can eventually determine the status of nodes and communication paths between them. A diagnostic algorithm for homogeneous systems (systems with only testing nodes) is given. The self-fault-diagnosis of inhomogeneous systems (systems with nodes of varying degrees of testing capability) is studied and diagnostic algorithms are proposed. >

Efficient distributed algorithms for self testing of multiple processor systems

Multiple processor systems allow both highly reliable and highly fast service. Distributed self-test algorithms that attempt to improve both the reliability and the performance of these systems are proposed. In these algorithms, reliability is improved by considering the distributed mode of control and assigning processors to test each other periodically for the diagnosis and isolation of the faulty processors and interprocessor links. Meanwhile, performance is improved by considering a dynamic testing strategy and minimizing testing overhead by reducing the number of tests performed on each processor. Simulation results show the effectiveness of the algorithms. >

A receiver-initiated load balancing method in computer networks using fuzzy logic control

We present a receiver-initiated fuzzy logic control method to improve computer network performance by balancing loads among computers. We study and present the performance of this protocol by comparing it with the performance of: (i) a well-known protocol called the BID algorithm; (ii) a non-fuzzy (threshold based) receiver-initiated algorithm; (iii) a fuzzy logic sender-initiated algorithm.

A methodology for evaluating load balancing algorithms

In general, a load balancing algorithm improves a system performance. Obviously, larger the difference between the task arrival rates at various processors, more the system is imbalanced and more improvement in the system performance is achieved using a load balancing algorithm. The existing works which have used an experimental technique to show the improvement in the system performance under a load balancing algorithm have used an ad hoc procedure to select the task arrival rates for various processors. Thus, their experimental results necessarily may not provide a complete picture of the improvement in the system performance under their load balancing algorithms. The authors present a systematic scheme for the selection of the task arrival rates at various processors such that experimental results reflect a complete picture of the improvement in the system performance under a load balancing algorithm. The idea has been motivated by the well-known Taguchi technique used in quality control.<<ETX>>

Scheduling routing table calculations to achieve fast convergence in OSPF protocol

Fast convergence to topology changes is now a key requirement in routing infrastructures while reducing the routing protocol’s processing overhead continues to be as important as before. In this paper, we examine the problem of scheduling routing table updates in link state routing protocols. Commercial routers typically use a hold time based scheme to limit the number of routing table updates as new LSAs arrive at the router. The hold time schemes limit the number of routing table updates at the expense of increased delay in convergence to the new topology, which is clearly not acceptable any more. We analyze the performance of different hold time schemes and propose a new approach to schedule routing table updates, called LSA Correlation. Rather than using individual LSAs as triggers for routing table updates, LSA Correlation scheme correlates the information in the LSAs to identify the topology change that led to their generation. A routing table update is performed when a topology change has been identified. The analysis and simulation results presented in this paper suggest that the LSA Correlation scheme performs much better than the hold time based schemes for both isolated and large scale topology change scenarios.

Stability analysis of a load balancing algorithm

Load balancing is the process of improving the performance of a system through a redistribution of loads among the processors. In this paper the authors present the stability analysis of a load balancing algorithm based on graph coloring. In this algorithm the processors use local knowledge for the purpose of load balancing. Unlike threshold based algorithms whose efficiency depends on the threshold level selected, the graph coloring based load balancing algorithm does not use any such global parameter. Furthermore, it uses the edge coloring concept to pair the processors. This avoids the selection/rejection operations encountered in many load balancing algorithms proposed in the literature. Also, there is no central controller and the algorithm is easily adaptable to changes in the system configurations.