waiting line systems

Upload: sudipitm

Post on 03-Apr-2018

218 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/29/2019 Waiting Line Systems

    1/57

    Waiting Line Systems

    Queuing Models

  • 7/29/2019 Waiting Line Systems

    2/57

    Examples of Waiting Lines

    At Petrol Pumps, Restaurants, Malls,

    banks, post offices

    Planes awaiting clearance from controltowers, trucks waiting to load or unloadcargo, cabs at airports, buses enteringterminals

    Employees waiting to swipe/punch cardsfor entry/exit

  • 7/29/2019 Waiting Line Systems

    3/57

    DehradunExp

    FerozPurJanta Exp

    PaschimExpress

    FrontierMail

    JammuTawi Exp

    Any Train Booking at All Booking Counters Shorter Queue and Less Idle Servers

    Specific Service CountersIdle Server at some counter,Longer Queue at some counter

    Indian Railways

    25 YearsAgo

  • 7/29/2019 Waiting Line Systems

    4/57

    Bulk Cash> 50K

    PowerAdvantage

    AccountHolderIn Person

    ThirdParty

    Premier

    Segregation of Service and Service PriorityHongKong Bank (HSBC)

    Nationalized Banks Security is a Warm FeelingService is a .. Feeling

    Receipts and PaymentsAt different counters.Idle time is more

    Longer queues

  • 7/29/2019 Waiting Line Systems

    5/57

    EnterExit

    CashierChitle Bandhu,Sadashiv Peth, Pune

    CustomerCounter (server)

    Every customer gets a numbered token. Customerspends. Less time at each counter. Purchases are

    updated at counter. Waiting Time at cashier reduced

  • 7/29/2019 Waiting Line Systems

    6/57

    Service Exit

    ProcessingOrder

    Waiting LineArrivals

    CallingPopulation

    Major Elements of Waiting Line Systems

  • 7/29/2019 Waiting Line Systems

    7/57

    Goals of a Waiting Line System

    Queuing Analysis for the design of Systemcapacity

    Waiting line models are predictive models ofexpected behavior of a system in which waiting

    line forms Balance cost of providing customer service vs

    cost of customers waiting for service Design or redesign to satisfy desired

    specifications (Bank Manager wants maximumno of waiting customers = 7 in a bank, henceestimate the no of tellers required)

  • 7/29/2019 Waiting Line Systems

    8/57

    Calling Population

    Infinite or unlimited

    Probability of arrival isnot significantly

    influenced by the factthat some customersare waiting

    Open to general

    public (theatres,banks, post offices,restaurants, etc..)

    Finite or limited

    Systems have limitedaccess for service

    Limited no ofmachines, limited noof trucks

  • 7/29/2019 Waiting Line Systems

    9/57

    Customer Arrivals

    Discrete people arriving, trucks arriving,telephone calls

    Arrivals in single units or batches (bus load of

    passengers arriving at fast food joints, Theatreor cinema hall patrons)

    Distribution of arrivals follow the PoissonDistribution (ex : 4 cars per hour)

    The average time in between arrivals follows thenegative exponential distribution. Ex Timebetween arrivals is 15 mins

  • 7/29/2019 Waiting Line Systems

    10/57

    Exit

    Single Channel/Server, Single Line, Single Phase

    Exit

    Exit

    Single Channel/Server, Single Line, Multiple Phase

    Payment of College Fees

  • 7/29/2019 Waiting Line Systems

    11/57

    Exit

    Multiple Channel/Server, Single Line, Single Phase

    Exit

    Exit

    Multiple Channel/Server, Multiple Line, Single Phase

    Exit

    Exit

    Exit

    Customer Service at Banks

    Hospitals Out Patient Department

  • 7/29/2019 Waiting Line Systems

    12/57

    Exit

    Multiple Channel/Server, Single Line, Multiple Phase

    Exit

    Exit

    Multiple Channel/Server, Multiple Line, Multiple Phase

    Exit

    Exit

    Exit

  • 7/29/2019 Waiting Line Systems

    13/57

    Poisson Distribution Assumptions

    Probability of occurrence of an event (arrival) ina given interval does not affect the probability ofoccurrence of an event in another nonoverlapping interval (afternoon & night show)

    The expected no of occurrences of an event(arrival) in an interval is proportional to the sizeof the interval (15 mins and 30 mins).

    Probability of occurrence of an event (arrival) inone interval is equal to the probability ofoccurrence of the event in another equal-sizeinterval/

  • 7/29/2019 Waiting Line Systems

    14/57

    Probability in a Poisson Distribution

    P(X) = e-u uX / X!

    u = expected mean number of occurrences within agiven interval

    e = Eulers constant = 2.71828

    X = number of occurrences of an event in a given timeinterval

    P(X) = Probability of occurrence of event X

    Excel Formula = POISSON(X,u,True) for Cummulative

    Excel Formula = POISSON(X,u,False) for NonCummulative

  • 7/29/2019 Waiting Line Systems

    15/57

    Poisson Distribution Problem

    Machines arrive at a repair shop at the rate of 3per 20 minute period. (u = 3)

    What is the probability that there will be 6

    machines arriving in a 20 minute period?P(X = 6)

    What is the probability that there will be 12machines arriving in 1 hour? (u = 9 per hour),

    P(X = 12) What is the probability that there will be less

    than 3 machines arriving in a 20 minute period?(u = 3), P(X< 3) = P(X = 0) + P(X = 1) + P(X = 2)

  • 7/29/2019 Waiting Line Systems

    16/57

    Types of Customers

    Arriving customers mayrefuse to enter thesystem as there is a longwaiting line

    Customers may arrive,wait for some time, andleave without beingserved

    Customers may switchlines in order to reducethe waiting time

    Balking

    Reneging

    Jockeying

  • 7/29/2019 Waiting Line Systems

    17/57

    Processing Order (Priority)

    First come, First served

    Assign priority and process waitingcustomers as per priority order

    High and Low Priority Customers

  • 7/29/2019 Waiting Line Systems

    18/57

    Service

    Number of servers

    Number of steps or

    phases

    Distribution of processingor service time

    Single or multiple

    Single phase or multiple

    phase

    Negative exponentialdistribution

    Most customers requireshort service time, while afew require more servicetime

  • 7/29/2019 Waiting Line Systems

    19/57

    Probability in a Exponential

    Distribution

    Probability density function f(t) = e- t for

    t, >= 0 and f(t) = 0 elsewhere

    = mean no of occurrences of a particularevent per time unit

    Mean u = Std Dev S = 1/

    P(t=b) = e- b = 1- EXPONDIST(b, ,True)

    P(a

  • 7/29/2019 Waiting Line Systems

    20/57

    Exponential Distribution Problem

    At a grocery store, time between arrivals is

    an exponential distribution. On anaverage, 2 customers arrive every four

    minutes. arr = 2/4 = 0.5 per min

    Service time is also exponentiallydistributed with a mean service rate of 40

    customers per hour. ser = 40/60 = 0.67per min

  • 7/29/2019 Waiting Line Systems

    21/57

    Exponential Distribution Problem

    Probability that no more than 4mins elapse between successivearrivals of customers

    Probability that more than 6 minselapse between successive arrivalsof customers

    Probability that between 4 and 6

    mins elapse between successivearrivals of customers Expected time (u) between

    successive arrivals and standarddeviation (s) of time betweenarrivals of customers

    Expected time of service Std deviation of service time Probability that service time is 1

    min or less Probability that service time is 2

    min or more Probability that service time is

    between 1 min and 2 min

    P(t=6) = e- (0.5)(6) =0.049787

    P(4

  • 7/29/2019 Waiting Line Systems

    22/57

    In the WaitingLine

    BeingServed

    In the System

    AverageNumber

    Lq /u L=Lq +/u

    AverageTime

    Wq =Lq/u 1/u W=Wq +1/u

    -------------System -----------

    Average Number Waiting and Average Waiting and Service Time

    Where = mean arrival rate and u = mean service rate

  • 7/29/2019 Waiting Line Systems

    23/57

    Lq The average number waiting for service

    L The average number in the system (waiting forservice or being served)

    P0The probability of zero units in the system

    p The system utilization (percentage of time serversare busy serving customers)

    WqThe average time customers wait for service

    WfThe average time customers spend in the system(waiting for service and service time)

    M The expected maximum number waiting forservice for a given level of confidence

    Measures of System Performance

  • 7/29/2019 Waiting Line Systems

    24/57

    Waiting Line Problem A car wash franchise estimates for a new proposed facility: car

    arrival rate = 20 cars per hour and car service rate u = 25 cars perhour. Service time is variable as cars are hand washed. Car arrivalsfollow Poisson distribution, while Car service time followsexponential distribution. Cars are processed one at a time. (Singleline, single server (s = 1)) Determine the following:

    Average no of cars being washed ( =/u = 20/25 = 0.80 car)

    Average no of cars in the system (cars being washed or waiting tobe washed, where average number waiting in line Lq = 3.2) (L=Lq+/u= 3.2 + 0.80 = 4.0 cars)

    Average time in the line (avg time cars wait to get washed) (=Lq/u =3.2/20 = 0.16 hour = 0.16*60 = 9.6 mins)

    Average time cars spend in the system (waiting in line and beingwashed) (= Wq +1/u = 9.6 + 1/25 = 0.20 hour = 0.20*60 = 12 mins)

    System utilization ( =/(su) = 20/(1*25) = 0.80 or 80%) where (s =no of servers = 1)

  • 7/29/2019 Waiting Line Systems

    25/57

    Basic Single Channel (M/M/1) Model

    M/M/1 = Markovian (Poisson) arrivals,Markovial (negative exponential) service, and1 (single) server

    One server or channel

    A Poisson arrival rate

    A negative exponential service time

    First-come, First-served processing order

    An infinite calling population

    No limit on queue length

  • 7/29/2019 Waiting Line Systems

    26/57

    (/u)Wa + (1 - /u)0(/u) + (1 - /u)

    = (/u)WaWq =

    Wq = Weighted Average of waiting time of ActualWaiting and Non waiting

    Wa = Average Waiting TimeProportion of Customers Waiting = (/u)

    Proportion of Customers who do not wait = (1 - /u)

    Average Waiting Time

  • 7/29/2019 Waiting Line Systems

    27/57

    Performance Measure Notation Formulae

    System utilization p /u

    Average number in line Lq 2/u(u- )

    Average number in line L Lq + /u

    Average time in Line Wq Lq / = /(u(-u) = (/u)Wa

    Average time in System W Wq + 1/u

    Probability of Zero units in System P0 (1 - /u)

    Probability of n units in System Pn P0(

    /u)

    n

    Probability that waiting line wont exceed kunits

    Pn

  • 7/29/2019 Waiting Line Systems

    28/57

    M/M/1 Problem

    At a ticket counter, mean arrival rate of

    customers = 3 per minute. Mean servicerate is 4 customers per minute. Calculate

    all the performance measures.

    Consider n = 2 customers in the systemand k = maximum length of the queue = 5

  • 7/29/2019 Waiting Line Systems

    29/57

    Performance Measure Notation Formulae

    System utilization p /u = = 0.75 or 75%

    Average number in line Lq 2/u(u- ) = 2.25 customers

    Average number in line L Lq + /u = 2.25 + = 3 customers

    Average time in Line Wq Lq / = 2.25/3 = 0.75 minute

    Average time in System W Wq + 1/u = 0.75 + = 1.0 minute

    Probability of Zero units in System P0 (1 - /u) = (1 0.75) = 0.25

    Probability of n units in System Pn P0(/u)n = 0.25(0.75)2 = 0.1406

    Probability that waiting line wontexceed k units

    Pn

  • 7/29/2019 Waiting Line Systems

    30/57

    Basic Multiple Channel (M/M/S) Model

    M/M/S = Markovian (Poisson) arrivals,

    Markovial (negative exponential) service, andS (multiple) server (S > 1)

    More than One server or channel

    A Poisson arrival rate A negative exponential service time

    First-come, First-served processing order

    An infinite calling population No upper limit on queue length

    The same mean service rate for all servers

  • 7/29/2019 Waiting Line Systems

    31/57

    Multiple Server, Priority Servicing

    Model

    Basic Multiple Server Model but there is

    priority servicing.

    Within each class (or priority), waiting unitsare processed in the order they arrive.

    Unit with low priority will have a longwaiting line

  • 7/29/2019 Waiting Line Systems

    32/57

    No Priority

    Priority 1

    Priority 2

    Priority 3

    New arrivals

    Multiple Server, Priority Servicing Model

  • 7/29/2019 Waiting Line Systems

    33/57

    Psychology of Waiting

    Perception of waitingof the customer

    Desirable waiting timeexploited

    Magazines providedat doctors or dentistwaiting rooms

    Music, in-flight movies Fill up forms

    Mirrors near lifts

    Impulse purchases atsupermarkets

    Additional spending

  • 7/29/2019 Waiting Line Systems

    34/57

    Value of Waiting Line Models

    Assumptions are criticized as :

    Often service times are not negativeexponential

    System is not in steady-state, but tends tobe dynamic

    Service is difficult to define because

    service requirements can varyconsiderably from customer to customer

  • 7/29/2019 Waiting Line Systems

    35/57

    OTHER WAITING LINEMODELS

  • 7/29/2019 Waiting Line Systems

    36/57

    First Come First Served processing

    in M/M/S Model

    Customers wait in a single line (post offices).

    Others systems record order of arrival (busyrestaurants) or have customers take a number

    on arrival (bakeries, customer service in banks) Supermarket checkouts do not fall in multiple

    servers (though they have multiple servers) ascustomers do not form a single line)

    Condition su > must be satisfied

    B i M l i l S S M/M/S

  • 7/29/2019 Waiting Line Systems

    37/57

    Performance Measure Notation Formulae

    System utilization p /su

    Average number in line Lq u(/u)sP0/((s-1)!(su- )2)

    Average number in line L Lq + /u

    Average time in Line Wq Lq / = /(u(-u) = (/u)Wa

    Average time in System W Wq + 1/u

    Probability of Zero units in System P0 1/((/u)n/n!) + (/u)s/(s!(1-(/su))

    from n = 0 to s-1

    Probability of n units in System Pn P0(/u)n/(n!) for n s

    Probability that an arrival will have to wait forservice

    Pw (/u)s P0/ (s!(1- /su))

    Average waiting time for an arrival not servedimmediately

    Wa 1/(su- )

    Basic Multiple Server System M/M/S

  • 7/29/2019 Waiting Line Systems

    38/57

    /u s Lq P0

    0.25 1 0.083 0.750

    2 0.004 0.778

    0.50 1 0.5 0.52 0.033 0.6

    3 0.003 0.606

    0.75 1 2.25 0.25

    2 0.123 0.455

    3 0.015 0.471

    1.0 2 0.333 0.333

    3 0.045 0.364

    4 0.007 0.367

    Infinite Source values for Lq and Po given /u and s

    And so on Refer Excel Worksheet

  • 7/29/2019 Waiting Line Systems

    39/57

    Determining Maximum Length of

    Waiting Lines

    Amount of space required toaccommodate waiting customers

    Number of customers waiting will not

    exceed a specified percentage of time Determine the line length (n) that probably

    will not exceed 95% or 99% of the time

    pn

    = K where p = /su K = (1 probability)/Lq (1- p)

    n = LogK/Logp

  • 7/29/2019 Waiting Line Systems

    40/57

    Problem

    Determine the maximum number of customers waiting inline for probabilities of both 95% and 99%. s = 1, = 4per hour, u = 5 per hour

    p = /su = 4/(1)(5) = 0.80

    For /u = 0.80, s = 1, Lq = 3.2

    For 95%, K = (1-0.95)/3.2(1-0.80) = 0.078

    n = log 0.078/log0.80 = 11.43 or 12

    For 99%, K = (1-0.99)/3.2(1-0.80) = 0.0156

    n = log 0.0156/log0.80 = 18.64 or 19

  • 7/29/2019 Waiting Line Systems

    41/57

    During noon hour at a bank, the arroval rate is 8 customers per minute,and the service rate is 2 customers per minute. Thus /u = 8/2 = 4.

    How many servers would be required so that the line does not exceed 4 waiting

    Customers? 6 servers at 95%, 7 servers at 99%

    Servers(s)

    Lqp = /su n95 n99

    5 2.216 0.80 10 17

    6 0.570 0.67 4 8

    7 0.180 0.57 1 4

    8 0.059 0.50 0

  • 7/29/2019 Waiting Line Systems

    42/57

    Cost Considerations

    Minimize Total Cost = Customer Waiting

    Cost + Capacity (Service) Cost

    Iterative procedure : Increase Capacity (orchannels) by one. And compute the totalcost.

    Total cost will initially decrease with

    increase in number of channels and thenincrease.

  • 7/29/2019 Waiting Line Systems

    43/57

    Trucks arrive at a warehouse at a rate of 15 per hour during business hoursCrew can unload at a rate of 5 per hour. Crew wage rates have increased andCrew size is to be determined. Crew cost is Rs 100/hour, while driver cost isRs 120 per hour. Thus /u = 15/5 = 3. Values of Lq are as per calculations

    for given s and /u

    Crew Size (s) Service Cost

    Rs 100 x s

    Avg No inSystem L = Lq

    + /u

    Waiting Cost= L x Rs 120/-

    Total Cost =Service Cost

    + Waiting

    Cost4 Rs 400 1.528 + 3 =

    4.5284.528 x 120 =

    Rs 543.26Rs 943.26

    5 Rs 500 0.354 + 3 =3.354

    3.354 x 120 =Rs 402.58

    Rs 902.58(minimum)

    6 Rs 600 0.099 + 3 =3.099

    3.099 x 120 =Rs 371.88

    Rs 971.88

    7 Rs 700 0.028 + 3 =3.028

    3.028 x 120 =Rs 363.36

    Rs 1063.36

  • 7/29/2019 Waiting Line Systems

    44/57

    Other Queuing Models

    Poisson Arrival Rate with anyservice Distribution

    Poisson Arrival Rate and ConstantService Time

    Finite Queue Length

    Finite Calling Population

    Multiple Server, Priority Servicing

    Model

    M/G/1 (single Server)

    M/D/1 (single server,Deterministic service time

  • 7/29/2019 Waiting Line Systems

    45/57

    Basic Single Channel (M/G/1) Model

    M/M/1 = Markovian (Poisson) arrivals,General service distribution, and 1 (single)server

    One server or channel

    A Poisson arrival rate Service time follows any distribution

    Estimate of variance is necessary

    First-come, First-served processing order

    An infinite calling population

    No limit on queue length

  • 7/29/2019 Waiting Line Systems

    46/57

    PerformanceMeasure Formula

    System Utilization p /su (where s = 1)

    Average number

    waiting in line

    Lq ((/u)2 +2v)/2(1-/u)

    where v = variance

    Average number inSystem

    L Lq + /u

    Average time waitingin line

    Wq Lq /

    Average time in thesystem

    W Wq + 1/u

    M/G/1 Poisson Arrival rate with any Service Distribution

  • 7/29/2019 Waiting Line Systems

    47/57

    Problem Any Service Distribution

    Customer arrival rate is Poisson = 0.25/min

    Service time has mean = 2 min (u = = 0.5)and standard deviation = 0.9 min (variance =

    0.81). Compute performance measures

  • 7/29/2019 Waiting Line Systems

    48/57

    Performance Measure Formula

    System Utilization p /su (where s = 1) = 0.25/0.50 = 0.50

    Average number

    waiting in line

    Lq ((/u)2 +2v)/2(1-/u) where v = variance

    = (0.25/0.50)2

    + 0,252

    (0.81)/2(1-(0.25/0.5)) =0.301 customers

    Average number inSystem

    L Lq + /u = 0.301 + 0.5 = 0.801 customers

    Average time waiting in

    line

    Wq Lq / = 0.301/0.25 = 1.203 min

    Average time in thesystem

    W Wq + 1/u = 1.203 + 1/0.5 = 3.203 min

    M/G/1 Poisson Arrival rate with any Service Distribution

  • 7/29/2019 Waiting Line Systems

    49/57

    Basic Single Channel (M/D/1) Model

    M/M/1 = Markovian (Poisson) arrivals,Constant Service time, and 1 (single) server

    One server or channel

    A Poisson arrival rate

    Service time is constant

    Variance = zero

    First-come, First-served processing order

    An infinite calling population

    No limit on queue length

  • 7/29/2019 Waiting Line Systems

    50/57

    Performance Measure Formula

    System Utilization p /su (where s = 1)

    Average number

    waiting in line

    Lq (/u)2/2(1-/u) = ()2/2u(u-)

    Average number inSystem

    L Lq + /u

    Average time waiting inline

    Wq Lq /

    Average time in thesystem

    W Wq + 1/u

    M/D/1 Poisson Arrival rate with Constant Service Time

  • 7/29/2019 Waiting Line Systems

    51/57

    Problem Constant Service Time

    Customer arrival rate is Poisson = 12/hour

    Service time has mean = 2 min (u = = 0.5)and standard deviation = 0.9 min (variance =

    0.81). Compute performance measures

  • 7/29/2019 Waiting Line Systems

    52/57

    Performance Measure Formula

    System Utilization p /su (where s = 1) = 12/20 = 0.60

    Average number

    waiting in line

    Lq 2/2u(u-) = 122/2(20)(20-12) = 0.45 customer

    Average number inSystem

    L Lq + /u = 0.45 + 0.6 = 1.05 customers

    Average time waiting inline

    Wq Lq / = 0.45/12 = 0.0375 hour = 2.25 min

    Average time in thesystem

    W Wq + 1/u = 2.25 + 60(1/20) = 5.25 min

    M/G/1 Poisson Arrival rate with Constant Service Time

    B i Si l Ch l ith Fi it Q

  • 7/29/2019 Waiting Line Systems

    53/57

    Basic Single Channel with Finite QueueLength Model

    M/M/1 = Markovian (Poisson) arrivals,Markovian Exponential Distribution Servicetime, and 1 (single) server

    One server or channel

    A Poisson arrival rate Exponential Distribution Service rate

    Limit on Maximum length of Queue First-come, First-served processing order

    An infinite calling population New customers not allowed when queue ismaximum (parking lots)

    Basic Single Server System M/M/1

  • 7/29/2019 Waiting Line Systems

    54/57

    Performance Measure Notation Formulae

    System utilization p /u

    Average number in line Lq L (1-P0)

    Average number in system L (/u)/(1- /u) (m+1)(/u)m+1/(1- (/u)m+1)

    Average time in System W Lq /((1-Pm)) + 1/u

    Average time in Line Wq W- 1/u

    Probability of Zero units in System P0 (1 - /u)/(1-(/u)m+1)where m = max length of queue

    Probability of n units in System Pn P0(/u)n

    Basic Single Server System M/M/1

  • 7/29/2019 Waiting Line Systems

    55/57

    Problem Finite Length

    Arrival rate = 9 per hour

    Service rate = 15 per hour

    Max length of Queue = 5 customers

    Calculate performance measures

    Basic Single Server System M/M/1 with Finite Queue Length

  • 7/29/2019 Waiting Line Systems

    56/57

    Performance Measure Notation Formulae

    System utilization p /u = 9/15 = 0.60

    Average number in line Lq L (1-P0) = 1.206 (1-0.420) = 0.63

    Average number in system L (/u)/(1- /u) (m+1)(/u)m+1/(1- (/u)m+1)

    = 0.6/(1-0.6) (5+1)(0.6)5+1)/(1-0.65+1) =1.206

    Average time in System W Lq /((1-Pm)) + 1/u = 0.63/(9(1 0.033)) +1/15 = 0.139 hr

    Average time in Line Wq W- 1/u = 0.139 1/15 = 0. 072 hour

    Probability of Zero units in System P0

    (1 - /u)/(1-(/u)m+1) = (1-0.6)/(1-0.65+1) = 0.420

    Probability of n units in System Pn P0(/u)n = 0.420(0.6)5 = 0.033

    Maximum length of Queue m Given = 5

    Basic Single Server System M/M/1 with Finite Queue Length

    Basic Single Channel with Finite Calling

  • 7/29/2019 Waiting Line Systems

    57/57

    Basic Single Channel with Finite CallingPopulation Model

    M/M/1 = Markovian (Poisson) arrivals,Markovian Exponential Distribution Servicetime, and 1 (single) server

    One server or channel

    A Poisson arrival rate Exponential Distribution Service rate

    First-come, First-served processing order

    A Finite calling population

    Machine Operator responsible for loadingand unloading five machines