## Fast Simulation of Steady-State Availability in Non-Markovian Systems (1992)

Citations: | 12 - 3 self |

### BibTeX

@MISC{Nicola92fastsimulation,

author = {Victor F. Nicola and Perwez Shahabuddin and Philip Heidelberger and Peter W. Glynn},

title = {Fast Simulation of Steady-State Availability in Non-Markovian Systems},

year = {1992}

}

### OpenURL

### Abstract

. This paper considers efficient simulation techniques for estimating steady-state quantities in models of highly dependable computing systems with general component failure and repair time distributions. Earlier approaches in this application setting for steady-state estimation rely on the regenerative method of simulation, which can be used when the failure time distributions are exponentially distributed. However, when the failure times are generally distributed the regenerative structure is lost and a new approach must be taken. The approach we take is to exploit a ratio representation for steady-state quantities in terms of cycles that are no longer independent and identically distributed. A "splitting" technique is used in which importance sampling is used to speed up the simulation of rare system failure events during a cycle, and standard simulation is used to estimate the expected cycle length. Experimental results show that the method is effective in practice. Keywords. Relia...

### Citations

461 | Monte Carlo Methods - Hammersley, Handscornb - 1964 |

113 | Importance sampling for stochastic simulations - Glynn, Iglehart - 1989 |

74 |
A quick simulation method for excessive backlogs in networks of queues
- Parekh, Walrand
- 1989
(Show Context)
Citation Context ...e new estimator willsbe much less than the variance of the standard estimator.sImportance sampling has been effectively employed in asvariety of situations, including queueing models (see, e.g.,s[9], =-=[26]-=- and [27]). Another approach to variance reductionswhen estimating long-run averages in models of communi-scation networks is considered in [22].sIts use in simulating highly dependable systems of the... |

73 | A GSMP formalism for discrete event systems
- Glynn
- 1989
(Show Context)
Citation Context ...s in Non-Regenerative SystemssConsider systems that can be modelled as (time homo-seneous Generalized Semi-Markov Processes (GSMP’s).s$or a d etaded . exposition on GSMP’s the reader is re-sferred to =-=[ll]-=- see also [25] for an alternative description).sterized by an output state vector, say X( t ) ,sthat takessvalue in Z‘, and an internal state vector, say S(t), thatstakes values in R” (Z G set of inte... |

71 |
Probability
- Breiman
- 1968
(Show Context)
Citation Context ...en { X ( t )s: t 1 0)senters A. Then under fairly general recurrence type con-sditions (which also ensure that the system returns to statesA infinitely often) we have that (see [5] and Section 6.9 ofs=-=[2]-=- for details; see also [7])sState MeasuressLet C be the length of an A - cycle.s39sAuthorized licensed use limited to: Stanford University. Downloaded on August 03,2010 at 00:29:49 UTC from IEEE Xplor... |

64 |
A unified framework for simulating Markovian models of highly reliable systems
- Goyal, Shahabuddin, et al.
- 1992
(Show Context)
Citation Context ...od issshown to be highly effective in practice. In addition, if thesfailure and repair distributions are exponential, then thesapproach is closely related to using “balanced failure bias-sing” ([29], =-=[15]-=-) and “measure specific dynatmic importancessampling” (MSDIS) [15] which, for Markovian models, wassshown to have bounded relative error for estimating thessteady-state unavailability in [29 . Also, w... |

53 |
A Guide to Simulation, Second Edition
- Bratley, Fox, et al.
- 1987
(Show Context)
Citation Context ...6)% confidence interval (CI) for U. The halfswidth (HW) of this CI will be given by 2 6 / 2 ( T / f i ,swheres2 6 / 2sis the l O O ( 1s- 612) percentile point of the standardsnormal distribution (see =-=[l]-=-). If the Di’s were i.i.d., thensu2 = Var(D;) which would be easy to estimate. If wesdiscard some of the initial Di’s (i.e., allow the system tosreach steady-state), then the Di’s from then on are (ap... |

53 |
Importance sampling for the simulation of highly reliable Markovian systems
- Shahabuddin
- 1994
(Show Context)
Citation Context ...e method issshown to be highly effective in practice. In addition, if thesfailure and repair distributions are exponential, then thesapproach is closely related to using “balanced failure bias-sing” (=-=[29]-=-, [15]) and “measure specific dynatmic importancessampling” (MSDIS) [15] which, for Markovian models, wassshown to have bounded relative error for estimating thessteady-state unavailability in [29 . A... |

31 | Monte Carlo simulation of Markov unreliability models - Lewis, Bohm - 1984 |

28 |
Optimally efficient estimation of the statistics of rare events in queueing networks
- Frater, Lenon, et al.
- 1991
(Show Context)
Citation Context ...of the new estimator willsbe much less than the variance of the standard estimator.sImportance sampling has been effectively employed in asvariety of situations, including queueing models (see, e.g.,s=-=[9]-=-, [26] and [27]). Another approach to variance reductionswhen estimating long-run averages in models of communi-scation networks is considered in [22].sIts use in simulating highly dependable systems ... |

27 | Using CSIM to model complex systems - Schwetman |

24 | Simulating stable stochastic systems, iii: Regenerative processes and dis crete-event simulation - Crane, Iglehart - 1975 |

23 |
Bounded relative error in estimating transient measures of highly dependable nonMarkovian systems
- Heidelberger, Shahabuddin, et al.
(Show Context)
Citation Context ...ormation ofsconfidence intervals.sWhile efficient simulation techniques for estimatingstransient quantities in models with generally distributedsfailure and repair times have been considered in [17], =-=[18]-=-,s[25] and 241, until now, these techniques have not beensdue to the fact that an entrance to the “all componentssup” state no longer constitutes a regeneration. In thisspaper, we show how these techn... |

22 |
Markov chains as an aid in the study of Markov processes
- Jensen
- 1953
(Show Context)
Citation Context ...experiment with an increasing failuresrate Weibull distribution, as will be described in Sections4.3.s3.1 A Uniformization-Based ImportancesThe uniformization technique (also known as random-sization =-=[19]-=-) can be used to sample from general distri-sbutions t .g. ,snonhomogeneous Poisson processes) withsbounded azard rate functions. Such distributions includesMarkovian phase-type, but exclude discrete,... |

22 | Fast simulation of dependability models with general failure, repair and maintenance processes - Nicola, Nakayama, et al. - 1990 |

20 |
Modeling and analysis of computer system availability
- Goyal, Lavenberg
- 1987
(Show Context)
Citation Context ... approach to variance reductionswhen estimating long-run averages in models of communi-scation networks is considered in [22].sIts use in simulating highly dependable systems of thestype described in =-=[14]-=- has been studied in [3], [4], [lo],s*Current address: Department of Com u t usScience, Uni-sversity of Twente, P.O. Box 217, 7500 XE Enschede, ThesNetherlands.sPeter W. GlynnsDepartment of Operations... |

17 |
Large deviations and efficient simulation of excessive backlogs in a GI/G/m queue
- Sadowsky
- 1991
(Show Context)
Citation Context ...imator willsbe much less than the variance of the standard estimator.sImportance sampling has been effectively employed in asvariety of situations, including queueing models (see, e.g.,s[9], [26] and =-=[27]-=-). Another approach to variance reductionswhen estimating long-run averages in models of communi-scation networks is considered in [22].sIts use in simulating highly dependable systems of thestype des... |

14 |
Ultrahigh reliability estimates through simulation
- Geist, Smotherman
- 1989
(Show Context)
Citation Context ...n estimating long-run averages in models of communi-scation networks is considered in [22].sIts use in simulating highly dependable systems of thestype described in [14] has been studied in [3], [4], =-=[lo]-=-,s*Current address: Department of Com u t usScience, Uni-sversity of Twente, P.O. Box 217, 7500 XE Enschede, ThesNetherlands.sPeter W. GlynnsDepartment of Operations ResearchsStanford University, CA 9... |

14 |
Uniformization and exponential transformation: Techniques for fast simulation of highly dependable non-Markovian systems
- Nicola, Heidelberger, et al.
- 1992
(Show Context)
Citation Context ... event oC the simulatedsnonhomogeneous Poisson process.sIn the context of reliability estimation, uniformizationscan be used to implement importance saimpling in Non-sMarkovian models as described in =-=[24]-=-. In a similar way,sit can also be used for unavailability estim4ation as we willsdescribe in this section.sConsider a system with N components. At any time t ,slet O(t) be the set of operational comp... |

13 |
Fast transient simulation of Markovian models of highly dependable systems. Performance Evaluation 20:267–286
- Shahabuddin
- 1994
(Show Context)
Citation Context ... fail with probability p,,; = l/IO(T,,)l. In addi-stion to being simple and robust, “balanced fadure biasing”sis known to be provably effective in the context of unreli-sability estimation [MI, [29], =-=[30]-=-.sThe likelihood ratio is computed recursively, by updat-sing it only at pseudo and failure event times as follows.sLet L k ,sk = 0,1,2, ..., be the likelihood ratio at the le - ths(pseudo or failure)... |

12 |
Importance sampling for stochastic simulations. Management Science 35:1367–1392
- Glynn, Iglehart
- 1989
(Show Context)
Citation Context ...o accurately estimate U; typ-sically, the closer U is to zero, the longer a standard simu-slation needs to be run.sWe seek to avoid such long run lengths by using thestechnique of importance sampling =-=[13]-=-, [16]. In impor-stance sampling, the system is simulated using a new setsof input distributions (e.g., failure distributions) that areschosen in such a way as to make the rare event much moreslikely ... |

11 | Fast simulation of Markovian reliability/availability models with gen- eral repair policies - Juneja, Shahabuddin - 1992 |

8 |
Efficient transient simulation of failure/repair Markovian models
- Carrasco
- 1991
(Show Context)
Citation Context ...nswhen estimating long-run averages in models of communi-scation networks is considered in [22].sIts use in simulating highly dependable systems of thestype described in [14] has been studied in [3], =-=[4]-=-, [lo],s*Current address: Department of Com u t usScience, Uni-sversity of Twente, P.O. Box 217, 7500 XE Enschede, ThesNetherlands.sPeter W. GlynnsDepartment of Operations ResearchsStanford University... |

8 |
A uniform theory for sums of Markov chain transition probabilities. The Annals of Probability 3
- Cogburn
- 1975
(Show Context)
Citation Context ...( t ) )sat the times when { X ( t )s: t 1 0)senters A. Then under fairly general recurrence type con-sditions (which also ensure that the system returns to statesA infinitely often) we have that (see =-=[5]-=- and Section 6.9 ofs[2] for details; see also [7])sState MeasuressLet C be the length of an A - cycle.s39sAuthorized licensed use limited to: Stanford University. Downloaded on August 03,2010 at 00:29... |

8 |
Estimating time av- erages via randomly-spaced observations
- Fox, Glynn
- 1989
(Show Context)
Citation Context ...rs A. Then under fairly general recurrence type con-sditions (which also ensure that the system returns to statesA infinitely often) we have that (see [5] and Section 6.9 ofs[2] for details; see also =-=[7]-=-)sState MeasuressLet C be the length of an A - cycle.s39sAuthorized licensed use limited to: Stanford University. Downloaded on August 03,2010 at 00:29:49 UTC from IEEE Xplore.sRestrictions apply.sC w... |

7 | Simultaneous and efficient simulation of highly dependable systems with different underlying distributions - Heidelberger, Nicola, et al. - 1992 |

6 | Fault injection simulation: A variance reduction technique for systems with rare events
- Moorsel, Haverkort
- 1991
(Show Context)
Citation Context ... situations, including queueing models (see, e.g.,s[9], [26] and [27]). Another approach to variance reductionswhen estimating long-run averages in models of communi-scation networks is considered in =-=[22]-=-.sIts use in simulating highly dependable systems of thestype described in [14] has been studied in [3], [4], [lo],s*Current address: Department of Com u t usScience, Uni-sversity of Twente, P.O. Box ... |

4 |
Failure distance-based simulation of repairable fault-tolerant systems
- Carrasco
- 1991
(Show Context)
Citation Context ...uctionswhen estimating long-run averages in models of communi-scation networks is considered in [22].sIts use in simulating highly dependable systems of thestype described in [14] has been studied in =-=[3]-=-, [4], [lo],s*Current address: Department of Com u t usScience, Uni-sversity of Twente, P.O. Box 217, 7500 XE Enschede, ThesNetherlands.sPeter W. GlynnsDepartment of Operations ResearchsStanford Unive... |

4 | Replication schemes for limiting expectations - Fox, Glynn - 1989 |

2 | Simulation of Highly Reliable Markovian and Non-Markovian Systems - Nakayama - 1991 |

1 | Fast simulation for transient measures and their gradients in highly reliable Markovian systems - Shahabuddin, Nakayama - 1992 |

1 | Simulating stG ble stochastic systems, 111: Regenerative processes and discrete event simulations - Crane, Iglehart - 1975 |

1 |
Efficient estimation of steady state measures in non-regenerative dependability models. To a pear
- Glynn, Heidelberger, et al.
- 1993
(Show Context)
Citation Context ...s showed the metho'd to be effectivesin practice for estimating the steady-state unavailability.sFor the efficient estimation of the steady-state mean timesbetween failures, the reader is referred to =-=[12]-=-.sAs a result of this work, the class of highly depend-sable systems that can be efficiently simulated to estimatessteady-state measures is greatly broadened.s6 AcknowledgementssThe work of Peter W. G... |

1 |
Simultaneous and efficient simulation of highly dependable systems with different underlying distribu- tions
- Methuen, Co, et al.
- 1992
(Show Context)
Citation Context ...rately estimate U; typ-sically, the closer U is to zero, the longer a standard simu-slation needs to be run.sWe seek to avoid such long run lengths by using thestechnique of importance sampling [13], =-=[16]-=-. In impor-stance sampling, the system is simulated using a new setsof input distributions (e.g., failure distributions) that areschosen in such a way as to make the rare event much moreslikely to occ... |

1 | Monte Carlo simula- tion of Markov unreliability models. Nuclear Engineer- ing and Design 77: 49-62. Authorized licensed use limited to: Stanford University. Downloaded on August 03,2010 at 00:29:49 UTC from IEEE Xplore. Restrictions apply - Lewis, Bohm - 1984 |

1 | Simulation of Highly Reliable Markovzan and Non-Markovian Systems - Nakayama - 1991 |