Chapter 9 · Introduction

Chapter 9 — Reliability, Linear Programming and Queues

Three classical models of operations research and engineering statistics: how long components and systems last, how to allocate limited resources optimally, and how queues of customers, jobs or data packets behave in the long run.

In the 1950s the reliability of electronic equipment became a crisis. Military and early commercial systems contained thousands of vacuum tubes, and a single tube failure could put a whole radar set out of action, so these systems spent much of their time under repair. Engineers learned that "how long will it last?" has to be asked twice: once for a component — whose risk of failure may rise, fall or stay level with age — and once for a system, whose reliability depends on how the components are wired together. The same mathematics now designs aircraft hydraulics, data centres and pacemakers, and the same statistics estimates a mean life from a test that is stopped after a few failures. At about the same time George Dantzig's simplex method (1947) turned the planning problems of logistics, diets and refineries into linear programmes that could be solved routinely, and the telephone engineer A. K. Erlang's earlier work on waiting lines (from 1909) grew into queueing theory, the basis for sizing call centres, hospital wards, toll plazas and computer networks.

The three sections of this chapter share a down-to-earth goal: to turn a physical system — a machine, a factory, a service counter — into a model simple enough to analyse exactly. Section 9.1 models lifetimes by their hazard rates and systems by their structure functions, and estimates mean lives from censored tests. Section 9.2 models optimal allocation as a linear objective on a convex polyhedron, solved by walking from corner to corner, with duality giving prices and certificates of optimality. Section 9.3 models congestion by birth–death processes (§3.3), whose steady states give the classical formulas for queue lengths and waiting times, and adds the basic inventory models. Exponential laws, the memoryless property and the Poisson process from Chapters 2 and 3 run through all of it.

Hazard rate h(t)→IFR / DFR ageing→Series, parallel, k-of-n→Censored life tests→Linear programmes→Simplex & duality→Markovian queues→L = λW, inventory
🎯 By the end of this chapter you can…
  • move between the reliability function, density, hazard rate, cumulative hazard and mean residual life of a lifetime, and compute mean times to failure;
  • decide whether a lifetime is IFR or DFR, prove the standard facts about ageing classes (minima, mixtures, the exponential bound), and know the hazards of the exponential, Weibull and gamma laws;
  • compute the reliability and the MTTF of series, parallel, $k$-out-of-$n$, standby and bridge systems, and explain why redundancy belongs at component level;
  • write the likelihood of Type I, Type II and randomly censored data and estimate the mean of an exponential lifetime, with exact $\chi^2$ confidence intervals;
  • formulate linear programmes, solve them graphically and by the simplex method (with artificial variables), recognise degeneracy and unboundedness, and use duality and complementary slackness;
  • derive the steady-state behaviour of the M/M/1, M/M/1/K, M/M/c and M/M/c/K queues, use Little's law and the Pollaczek–Khinchine formula for M/G/1, and solve the EOQ inventory model.
Teaser — The Bathtub Curve: Is an Old Machine Better or Worse?

A machine's hazard rate (failures per year among machines still working) is the sum of three causes: early defects (a Weibull hazard with shape $\tfrac12$, falling), random shocks (constant $0.02$ per year) and wear-out (a Weibull hazard with shape $4$, rising). Their sum, the thick curve, is shaped like a bathtub. Move the age slider or press Play, and compare the risk of failing within the next year and the expected remaining life with those of a brand-new machine.

0.5
age t—
hazard h(t) per year—
P(survive to t)—
P(fail within a year | alive at t)—
mean residual life m(t)—

How this is examined

CSIR NET draws on this chapter mainly through short computations with one decisive idea. From reliability, Part B asks for a conditional survival probability from a given hazard rate, the hazard rate of a named law, the reliability or mean time to failure of a small series, parallel or $k$-out-of-$n$ system of exponential components, or the maximum likelihood estimate of a mean life from a Type I or Type II censored exponential test; Part C asks which statements about IFR and DFR laws, mixtures, minima and maxima, redundancy and censored-data estimators are true, and each false option is one of the traps in the ⚠️ box of §9.1. From linear programming, Part B asks for the optimal value or an optimal vertex of a small programme, the dual of a given programme or the number of basic feasible solutions; Part C tests the duality theorems, complementary slackness, degeneracy, unboundedness and infeasibility. From queues, Part B asks for $L$, $L_q$, $W$ or $W_q$ in an M/M/1 or M/M/c system, or the economic order quantity; Part C asks which steady-state statements hold (for example, when a stationary distribution exists, or how $W$ changes when servers are pooled). GATE ST examines the reliability and queueing material in the same style, often as numerical-answer questions.

Sections in this chapter

✅ Before you start

For §9.1 you need the tail formula $E[X]=\int_0^\infty P(X\gt x)\,dx$ (§1.3); the exponential, gamma and Weibull laws and the lack of memory of the exponential (§2.1); Jensen's inequality (§2.2); the Poisson process (§3.3); order statistics and Rényi's representation of exponential spacings (§4.1); maximum likelihood, sufficiency, completeness, the Lehmann–Scheffé theorem and the Cramér–Rao bound (§4.2); and pivotal confidence intervals with the upper-point notation $\chi^2_{m,\alpha}$ (§4.3). For §9.2 you need systems of linear equations, rank and basic linear algebra (the sister course CSIR NET Unit 1: Analysis and Linear Algebra). For §9.3 you need continuous-time Markov chains and birth–death processes with their steady-state equations (§3.3) and stationary distributions (§3.2).