Optimal redundancy allocation to maximize multi-state computer network reliability subject to correlated failures
Introduction
In modern society, computer networks are the main medium for data transmission. Accidents or repair may disrupt network operation, preventing the execution of normal operations. Preserving network stability is thus important to ensure uninterrupted service. Therefore, many system supervisors focus on system reliability assessment and maximization for the computer networks they operate. The issues underlying system reliability assessment can be traced to binary-state computer network reliability assessment. In a binary-state computer network, each edge denoting a transmission line, may either work or not. System reliability is defined as the probability that the source vertex is connected to the sink vertex [6], [16]. Reliability assessment can also be extended to multi-state computer networks (MSCNs), where each edge is composed of several physical lines and each physical line may provide a fixed capacity (i.e. bandwidth) or fail. Clearly, such MSCNs are more detailed than their binary state counterparts. In the multi-state systems, reliability is defined as the probability that a computer network can successfully transmit d units of data from source to sink [34], [18], [17].
Since system reliability is regarded as a performance index to evaluate the capability of a computer network, most studies discuss system reliability maximization for MSCNs from the perspective of network topology optimization [5]; [28] or component assignment optimization [19], [21]. Network topology optimization focuses on exploring the connections between edges and vertices to search for the optimal network topology that maximizes system reliability. According to Stevenson [30], topology optimization is often expensive since it involves designing or redesigning networks. Component assignment optimization considers a set of multi-state components available for allocation to edges with the goal of determining the optimal component assignment that maximizes system reliability. The advantage of component assignment optimization is that system reliability can be improved without altering the existing network topology. However, in such situations, each component's maximal redundancy is fixed (i.e. each component's redundancy level is given) such that an edge may provide surplus capacity after allocating a component.
Redundancy allocation is another way to maximize system reliability for computer networks. The concept of redundancy allocation was introduced by Misra and Sharma [24]. It requires that each edge consists of several binary-state components allocated in parallel to enhance system stability. In other words, redundancy allocation determines the redundancy level for each edge and has been applied in many fields. Huang et al. [12] adopted a redundancy technique to promote the reliability for an aircraft's multi-channel electrical power supply system. Tian et al. [31] presented an optimization model for a multi-state series-parallel system to jointly determine the optimal component state distribution and optimal redundancy for each stage. Azaron et al. [3] solved a multi-objective discrete reliability optimization problem in a cold-standby redundant system using a genetic algorithm (GA). Tian et al. [32] developed a practical approach for the joint reliability–redundancy optimization of multi-state series-parallel systems. The approach not only determines the optimal redundancy level for each parallel subsystem, but also finds the optimal values for the variables that affect the component state distributions in each subsystem. Liu et al. [22] presented an approach of joint redundancy and imperfect maintenance strategy optimization for multi-state systems. Feizollahi et al. [8] proposed a robust optimization framework to deal with uncertain component reliabilities in redundancy allocation problems for series-parallel systems. Roy et al. [29] discussed a multi-objective reliability redundancy allocation problem for the series-parallel system and found the optimum number of redundant components to maximize system reliability and minimize system cost with entropy as the constraint. These studies demonstrated that an advantage of redundancy allocation is that each edge's redundancy level is appropriately determined so that the edge has no surplus capacity as much as possible.
In this study, redundancy allocation of an edge determines the number of physical lines allocated to the edge, i.e. redundancy level. For example, an edge with redundancy level 2 combines 2 physical lines. Since each physical line possesses two states, the edge has three states (i.e. no physical line works, one of physical line works, and both physical lines work) following a binomial distribution. Most studies [18], [17], [19], [21] assume the physical lines allocated to an edge are statistically independent. However, failure of a physical line may cause other lines in the edge to partially or completely fail. For example, a disaster may cause multiple physical lines to fail, necessitating repair. During repair of a physical line, other lines belonging to the edge need to be partially or fully suspended. Such a phenomenon is called a correlated failure and should not be ignored in system reliability assessment and optimization [9], [10].
This paper determines the optimal redundancy allocation of a computer network to maximize system reliability subject to a budget, while considering the impact of correlated physical-line failures. To our knowledge, this problem has not been considered previously. Chambari et al. [4] illustrated that redundancy allocation optimization is a NP-hard problem. Thus, the problem considered here is also NP-hard. In order to solve the optimal redundancy allocation problem with correlated failures, a simulated annealing (SA) approach is proposed based on the conceptual approach of Chambari et al. [4]. SA was introduced by Metropolis et al. [23] and has been widely applied to combinatorial optimization problems [33]. SA stimulates the annealing process of solids to find a near optimal solution and has the advantage of avoiding convergence to local optimum. According to the strategy suggested by Dong et al. [7], we integrate 1-opt and 2-opt operations in the stage of neighborhood search of the proposed SA to enhance its ability to perform global search. Moreover, a correlated binomial distribution model developed by Fiondella and Zeephongsekul [10] is utilized to evaluate the probability distribution for each edge under redundancy allocation when failures are correlated, and the system reliability is evaluated in terms of minimal paths (MP). The purposes of this study are to propose and solve the system reliability redundancy allocation problem with correlated failures and to demonstrate the computational efficiency of the proposed SA by comparing it with GA, particle swarm optimization (PSO), and tabu search (TS) through case studies of four practical computer networks.
The remainder of the paper is organized as follows. Assumptions and problem formulation are detailed in Section 2. State distribution of a correlated physical line redundancy allocation is illustrated in Section 3. Multi-state computer network model associated with a redundancy allocation is constructed in Section 4. Section 5 describes the proposed SA. Numerical experiments of four practical computer networks are executed to compare the proposed SA with several popular soft computing algorithms in Section 6. The conclusions are given in Section 7, along with future research.
Section snippets
Assumptions and problem formulation
Let E={ei|1≤i≤le;n} be a set of edges, where ei denotes the ith edge and V be a set of vertices. Thus, a computer network is represented as (E, V). The length of ei is denoted by li for i=1, 2, …, n. Moreover, let k be the capacity of a single physical line. The probability that a physical line can successfully transmit data (i.e., it can provide the capacity k) is denoted by p. The cost to transmit per unit of length of a physical line is denoted by c. A redundancy allocation is represented as
State distribution of correlated physical line redundancy allocation
Before developing an MSCN model associated with a redundancy allocation Y, we describe an edge state distribution where the physical lines may experience correlated failure. Consider an edge ei with redundancy level yi and each physical line providing capacity k in the operational state. The reliability and unreliability of each physical line is characterized by the probabilities p and (1–p) respectively. When the failures of the physical lines are uncorrelated, the probability distribution of
Multi-state computer network model associated with a redundancy allocation
This section describes the relationship between the flow and capacity associated with a redundancy allocation to build an MSCN model. We then propose an algorithm to measure system reliability.
Simulated annealing
The concept of SA is adopted from the “annealing” process used in the metallurgical industry. Initially, a solid is heated, and then a slow cooling schedule is executed to allow crystallization to a minimal energy. During the initial execution of the SA, the temperature should be high enough such that the probability of dropping in the local optimum is low. As the execution of cooling schedule progresses, the SA gradually converges to a near-optimum solution. To implement the proposed SA to
Numerical experiments
In this section, the proposed SA is applied to four practical computer networks which are ARPA, OCT, TANET, and NSFNET. Fig. 5, Fig. 6, Fig. 7, Fig. 8 depict their network topologies. ARPA is composed of 9 edges, 4 vertices, and 13 MPs, OCT is composed of 29 edges, 24 vertices, and 9 MPs, TANET is composed of 30 edges, 27 vertices, and 6 MPs, and NSFNET is composed of 16 edges, 14 vertices, and 6 MPs. The length data of the four networks are given in Table 3, Table 4, Table 5, Table 6(a),
Conclusions
This study discusses a physical line redundancy allocation optimization problem to maximize system reliability subject to a budget, where the physical lines allocated to the same edge may experience correlated failure. Considering failure correlation is more realistic than the majority of previous studies and the redundancy allocation strategy to maximize system reliability is more convenient than topology redesign and is also more flexible than the component allocation. The proposed SA
Acknowledgment
This work was supported in part by the Ministry of Science and Technology, Taiwan [Grant no. MOST 104-2410-H-128-014].
References (37)
- et al.
A hybrid variable neighborhood search and simulated annealing algorithm to estimate the three parameters of the Weibull distribution
Expert Syst Appl
(2011) - et al.
Routing and capacity assignment in backbone communication networks
Comput Oper Res
(1997) - et al.
Multi-objective reliability optimization for dissimilar-unit cold-standby systems using a genetic algorithm
Comput Oper Res
(2009) - et al.
An efficient simulated annealing algorithm for the redundancy allocation problem with a choice of redundancy strategies
Reliab Eng Syst Saf
(2013) - et al.
A comparison of a neighborhood search technique for forest spatial harvest scheduling problems: a case study of the simulated annealing algorithm
For Ecol Manag
(2015) - et al.
A cooperative parallel tabu search algorithm for the quadratic assignment problem
Eur J Oper Res
(2009) - et al.
Using minimal cuts to optimize network reliability for a stochastic computer subject to assignment budget
Comput Oper Res
(2011) - et al.
Stochastic computer network under accuracy rate constraint from QoS viewpoint
Inf Sci
(2013) - et al.
System reliability maximization for a computer network by finding the optimal two-class allocation subject to budget
Appl Soft Comput
(2015) - et al.
Two metaheuristics for solving the reliability redundancy allocation problem to maximize mean time to failure of a series–parallel system
Sci Iran
(2013)