(复杂系统的性能评价与优化课件资料）Queueing_Theory_Fudamentals.pptVIP

Copyright by Yu-Chi Ho Queueing Theory Fundamentals Elements of Q-Network Notations and definitions M/M/1 queue and others Little’s Law Burke’s theorem Queueing Network fundamentals Product form equation QNA, MPX? Notations and Definitions A/B/m/K/M Comment By arranging Erlangian and Hyper-exponential staged server in parallel and in series, we can in principle approximate any service distribution. M/M/1 queue Performance measure Average number of customers in the system N = r/(1-r) Average time in system = 1/(m-l) Average waiting time W = rN/l Average time in system = average waiting time + average service time (1/m). About N(r) Average number of customers in the system N = r/(1-r) What happens if r??0? N ?0 What happens if r ?1? N ?+? The hocky stick tradeoff The impact of randomness Using the Pollaczek-Kinchin formula, we can get the performance measure for M/G/1 and M/D/1. Some other extensions Discouraged arrivals vs. Poisson distribution M/M/∞=always exists new server for new customer M/M/m=there are m servers in the service station M/M/1/K=finite storage M/M/m/m=loss systems M/M/1//M=finite customer population M/M/∞//M=finite customer population, infinite server numbers M/M/m/K/M=general case Little’s Law The last arrival within t will leave the system at (t+e). So a(t)=d(t+e). Thus N(t)=a(t)-d(t)=d(t+e)-d(t). In steady state, e=T (the average flow time through the queueing system is T) and departure rate=arrival rate=l. So N(t)=d(t+e)-d(t)=l(t+e)-l(t)=le=lT. The size of the shaded area: Manufacturing example of Little’s Law Version 1: 10 people, 10 machines, 10 operations/part, one part arrival /day Everyone working on the same operation on all ten parts, (e.g. on day 1, all the 10 people work on operation No.1 for the 10 parts; on day 2, all work on operation No.2 for the 10 parts, etc.) Suppose in each day the operation starts after the part arrives. Then in version 1, in day 1, there is 0 part in the queue; in day 2, there is 1 part; in day 2, t