Queuing Theory Jackson Networks Network Model Consider a simple 3 stage model where the output of 2 queues becomes the input process for a third Station A Station C Station B Network Model Station A Station C Station B Proposition 1: rate in = rate out, that is, if lA and lB are the input rates for station A and B respectively, then the input rate for station C is lA+lB. Proposition 2: exponential inter-arrivals at A and B provide exponential inter-arrivals at station C. Network Model Station A Station C Station B Instructor Derivation of the minimum of 2 exponentials. P{min(X1, X 2 ) t} el1t el2t e(l1l2 )t provided, X1, X 2 exponentially distributed Jackson Network Station 1 (1 1 j )l1 1 l1 2 s1 Station v l 1 j l1 j Station j lj . . 1 2 1 sj j l 2 sv (1 j )l Jackson Network • All external arrivals to each station must follow a Poisson process (exponential inter-arrivals) • All service times must be exponentially distributed • All queues must have unlimited capacity • When a job leaves a station, the probability that it will go to another station is independent of its past history and of the location of any other job 1. Calculate the input arrival rate to each station 2. Treat each station independently as an M/M/s queue Calculate Arrival Rates m li i kilk k 1 where li calculate arrival rate at station i i external arrival rate at station i ki probability a job from station k goes to station i Example Jobs submitted to a computer center must first pass through an input processor before arriving at the central processor. 80% of jobs get passed on the central processor and 20% of jobs are rejected. Of the jobs that pass through the central processor, 60% are returned to the customer and 40% are passed to the printer. Jobs arrive at the station at a rate of 10 per minute. Calculate the arrival rate to each station. Example .6 .2 10 Input .8 Central l1 1 11l1 21l2 31l3 l2 2 12 l1 22 l2 32 l3 l3 3 13l1 23l2 33l3 l1 10 0 0 0 10 l2 0 .8l1 0 0 8 l3 0 0 .4l2 0 3.2 .4 Printer Example (cont.) • For the computer center the processing times are 10 seconds for the input processor, 5 seconds for the central processor, and 70 seconds for the printer. Our task is to determine the number of parallel stations (multiple servers) to have at each station to balance workload. Example 10 Input 8 Central 3.2 Printer M / M / s1 M / M / s2 M / M / s3 l 10 6 l 8 12 l 3.2 6/ 7 For our initial try, we will solve for s1 = 2, s2 = 1, s3 = 4 10 Input M /M /2 l 10 6 Metrics Model l r po Wq Lq 8 Central 3.2 Printer M / M /1 l 8 M /M /4 l 3.2 12 6/ 7 Input M/M/2 10 6 0.833 0.091 3.78 0.378 Central M/M/1 8 12 0.667 0.333 1.33 0.167 Printer M/M/4 3.2 6/7 0.933 0.009 8.68 2.71 Metrics Model l r po Wq Lq Input M/M/2 10 6 0.833 0.091 3.78 0.378 Central M/M/1 8 12 0.667 0.333 1.33 0.167 Printer M/M/4 3.2 6/7 0.933 0.009 8.68 2.71 If these numbers are correct, clearly Lq and Wq indicate the bottleneck is at the printer station. We may wish to add a printer if speedy return of printouts is required. Secondarily, overall processing may be increased by adding another input processor. Job Shop Example • An electronics firm has 3 different products in a job shop environment. The job shop has six different machines with multiple machines at 5 of the 6 stations. Product Order rate Flow 1 30/month ABDF 2 10/month ABEF 3 20/month ACEF Summary information is on the network below. Job Shop Example A 60 Machine B s=2 = 22 Machine A s=3 = 25 Machine D s=3 = 11 Machine E s=2 = 23 Machine C s=1 = 29 Machine F s=4 = 20 Job Shop Example A 60 Machine B s=2 = 22 Machine A s=3 = 25 Machine D s=3 = 11 Machine E s=2 = 23 Machine F s=4 = 20 Machine C s=1 = 29 l A A AAl A BAlB CA lC DA lD EAlE FAlF lB B ABl A BBlB CB lC DB lD EB lE FBlF lC C AC l A BC lB CC lC DC lD EC lE FC lF lD D ADl A BDlB CD lC DD lD EDlE FDlF lE E AE l A BE lB CE lC DE lD EE lE FE lF lF F AF l A BF lB CF lC DF lD EF lE FF lF Job Shop Example A 60 Machine B s=2 = 22 Machine A s=3 = 25 Machine D s=3 = 11 Machine E s=2 = 23 Machine F s=4 = 20 Machine C s=1 = 29 l A 60 0 l A 0 lB 0lC 0lD 0lE 0lF 60 lB 0 (40 / 60)l A 0lB 0lC 0lD 0lE 0lF 40 lC 0 (20 / 60)l A 0lB 0lC 0lD 0lE 0lF 20 lD 0 0l A (30 / 40)lB 0lC 0lD 0lE 0lF 30 lE 0 0l A (10 / 40)lB (20 / 20)lC 0lD 0lE 0lF 30 lF 0 0l A 0lB 0lC (30 / 30)lD (30 / 30)lE 0lF 60 Job Shop Example A 60 Machine B s=2 = 22 Machine A s=3 = 25 Machine D s=3 = 11 Machine E s=2 = 23 Machine F s=4 = 20 Machine C s=1 = 29 Lets calculate metrics for Machine B M/M/2 model with lB 40 22 l 40 r 0.91 s 2(20) Use formulas in Fig. 16.5 p. 564 or use charts p o .04 r 0.91 L 10 .4 r 0.91 Repeat for Machines A, C, D, E, F Machine D s=3 = 11 Machine B s=2 = 22 A 60 Machine A s=3 = 25 Machine E s=2 = 23 Machine F s=4 = 20 Machine C s=1 = 29 Metrics Model l r po Wq Lq W L A M/M/3 60 25 0.800 0.06 0.043 2.589 0.083 4.989 B M/M/2 40 22 0.909 0.04 0.216 8.658 0.262 10.476 C M/M/1 20 29 0.690 0.31 0.077 1.533 0.111 2.222 D M/M/3 30 11 0.909 0.02 0.278 8.332 0.369 11.059 E M/M/2 30 23 0.652 0.21 0.032 0.965 0.076 2.27 F M/M/4 60 20 0.750 0.042 0.025 1.528 0.075 4.528 Interesting Application to Manufacturing Metrics Model l r po Wq Lq W L A M/M/3 60 25 0.800 0.06 0.043 2.589 0.083 4.989 B M/M/2 40 22 0.909 0.04 0.216 8.658 0.262 10.476 C M/M/1 20 29 0.690 0.31 0.077 1.533 0.111 2.222 D M/M/3 30 11 0.909 0.02 0.278 8.332 0.369 11.059 E M/M/2 30 23 0.652 0.21 0.032 0.965 0.076 2.27 F M/M/4 60 20 0.750 0.042 0.025 1.528 0.075 4.528 Note that lead time is just the time in the system which for product 1 which has sequence ABDF is W = total time in system = lead time = WA + WB + WD + WF = .789 WIP = work in process = parts per month x lead time = 30(.789) = 23.67 Repeat for Machines A, C, D, E, F A 60 Machine B s=2 = 22 Machine A s=3 = 25 Machine D s=3 = 11 Machine E s=2 = 23 Machine F s=4 = 20 Machine C s=1 = 29 Monthly Product Orders Sequence 1 30 ABDF 2 10 ABEF 3 20 ACEF Lead Time 0.789 0.496 0.345 Queue Time 0.562 0.316 0.177 WIP 23.67 4.96 6.9 Some Recommendations Metrics Model l r po Wq Lq W L A M/M/3 60 25 0.800 0.06 0.043 2.589 0.083 4.989 B M/M/2 40 22 0.909 0.04 0.216 8.658 0.262 10.476 C M/M/1 20 29 0.690 0.31 0.077 1.533 0.111 2.222 D M/M/3 30 11 0.909 0.02 0.278 8.332 0.369 11.059 E M/M/2 30 23 0.652 0.21 0.032 0.965 0.076 2.27 F M/M/4 60 20 0.750 0.042 0.025 1.528 0.075 4.528 If lead time is too slow (WIP too high), one possibility is to add additional machining to the bottleneck area. This occurs at stations D and B. Note that this assumes the cost of the machines is not a critical part of the decision. This may or may not need to be included in a final recommendation.