Professional Documents
Culture Documents
Little PDF
Little PDF
LITTLE’S LAW
Input: Continuous flow or discrete units (examples: granules of powder measured in tons,
tons of paper, number of customers, $1000’s).
System: Boundary is all that is required (very general, abstract).
Output: Same as input, call it throughput.
• System in steady state/in the long run (for example, over many cycles).
Quantities that are related via Little’s law (long-run averages, or time-averages):
Little’s Law L = λW
1
Little’s
Little’s LawLaw
forforRetail
Retail calls,
calls, May
May2002:
2002:USUS
Bank
Bank
λ: Throughput Rate, Retail, May 2002; US Bank
Arrivals to offered Retail Total
May2002
50000
45000
40000
Number of cases
35000
30000
25000
20000
15000
10000
5000
0
0 5 10 15 20 25 30
days
40
35
30
Means, Seconds
25
20
15
10
0
0 5 10 15 20 25 30
days
5
Average Number in Queue
0
0 5 10 15 20 25 30
days
Date 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
λ 38476 36144 37414 14194 7107 38587 33572 33220 33349 34009 14807 7141 41293 37653 36872 35266
W 5.6 3.3 8.0 10.0 6.1 2.6 1.9 1.9 1.8 3.8 10.5 4.8 3.5 3.4 6.8 4.2
λ*W 2.49 1.39 3.48 1.65 0.50 1.14 0.74 0.74 0.69 1.50 1.79 0.40 1.69 1.47 2.91 1.71
L 2.50 1.40 3.49 1.66 0.50 1.15 0.74 0.74 0.71 1.51 1.81 0.40 1.71 1.47 2.94 1.73
Date 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
λ 35338 15533 7530 40534 35493 34070 34005 32512 13100 5909 1558 43980 38163 38416 40284
W 5.4 27.6 8.1 3.0 2.1 2.1 3.2 3.5 2.3 6.4 37.0 6.5 2.7 2.5 3.2
λ*W 2.19 4.96 0.71 1.41 0.87 0.83 1.25 1.30 0.35 0.44 0.67 3.29 1.18 1.11 1.47
L 2.20 4.97 0.71 1.42 0.88 0.84 1.27 1.31 0.37 0.44 0.67 3.30 1.20 1.12 1.48
2
1
th
Little’s
Little’s LawLaw
forfor Retailcalls,
Retail calls, August
August 16 , 2001:
16th, US Bank
2001: US Bank
λ: Throughput Rate, Retail, August 16th, 2001; US Bank
Arrivals to queue Retail
16 August 2001
2500
2000
Number of cases
1500
1000
500
0
7:00 9:00 11:00 13:00 15:00 17:00 19:00 21:00 23:00
Time (Resolution 30 min.)
70
60
50
Means, Seconds
40
30
20
10
0
7:00 9:00 11:00 13:00 15:00 17:00 19:00 21:00 23:00
25
20
Average Number in Queue
15
10
0
7:00 9:00 11:00 13:00 15:00 17:00 19:00 21:00 23:00
time (resolution 30 min)
Time 7:00 7:30 8:00 8:30 9:00 9:30 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00
λ 443 639 987 1291 1998 2166 2278 2231 2158 2135 2000 1408 1311 1303 1323 1285 1340
W 1.7 3.2 1.2 1.5 2.4 2.8 2.4 2.6 2.0 1.3 1.3 0.8 1.0 1.0 0.8 0.8 1.5
λ*W 0.42 1.14 0.68 1.06 2.72 3.42 3.01 3.18 2.44 1.55 1.47 0.64 0.72 0.72 0.62 0.59 1.09
L 0.42 1.14 0.68 1.06 2.72 3.40 3.02 3.17 2.41 1.59 1.48 0.64 0.72 0.72 0.62 0.57 1.11
Time 15:30 16:00 16:30 17:00 17:30 18:00 18:30 19:00 19:30 20:00 20:30 21:00 21:30 22:00 22:30 23:00 23:30
λ 1258 1235 1157 942 788 752 803 619 485 437 421 386 336 311 274 251 193
W 3.5 3.6 15.8 4.2 2.4 4.9 51.9 10.0 3.5 1.7 1.3 2.1 3.3 1.4 2.0 14.3 32.6
λ*W 2.422 2.45 10.2 2.173 1.06 2.05 23.16 3.43 0.95 0.41 0.314 0.44 0.62 0.24 0.30 2.00 3.50
L 2.37 2.49 10.17 2.16 1.07 1.94 23.11 3.59 0.95 0.40 0.31 0.45 0.62 0.24 0.30 1.83 3.63
3
2
th
Little’s
Little’s LawLaw
forfor Privatecalls,
Private calls, May
May 44th,
, 2004: Israeli
2004: Telecom
Israeli Telecom
λ: Throughput Rate, Private, May 4th, 2004; Israeli Telecom
Arrivals to queue Private
4 May 2004
600
500
Number of cases
400
300
200
100
0
7:00 9:00 11:00 13:00 15:00 17:00 19:00 21:00 23:00
Time (Resolution 30 min.)
90
80
70
Means, Seconds
60
50
40
30
20
10
0
7:00 9:00 11:00 13:00 15:00 17:00 19:00 21:00 23:00
Time (Resolution 30 min.)
25
20
Average Number in Queue
15
10
0
7:00 9:00 11:00 13:00 15:00 17:00 19:00 21:00 23:00
time (resolution 30 min)
Time 7:00 7:30 8:00 8:30 9:00 9:30 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00
λ 84 133 245 341 368 417 397 447 429 474 505 455 513 451 426 418 437
W 0.5 0.4 0.8 47.9 48.8 61.5 11.3 55.6 40.0 81.9 15.8 15.3 32.6 10.8 31.3 2.9 22.5
λ*W 0.02 0.03 0.11 9.08 9.98 14.24 2.50 13.81 9.52 21.57 4.44 3.86 9.29 2.70 7.40 0.67 5.46
L 0.02 0.03 0.11 8.47 10.59 14.24 1.85 14.40 9.42 21.57 4.61 3.86 8.98 2.92 7.49 0.67 5.46
Time 15:30 16:00 16:30 17:00 17:30 18:00 18:30 19:00 19:30 20:00 20:30 21:00 21:30 22:00 22:30 23:00 23:30
λ 449 452 458 486 534 508 557 433 450 448 408 389 347 285 274 208 147
W 7.2 5.1 17.1 25.8 15.6 18.9 35.8 3.0 18.3 59.5 24.1 47.7 23.6 32.5 65.3 59.5 1.3
λ*W 1.78 1.28 4.36 6.96 4.62 5.34 11.07 0.723 4.572 14.8 5.45 10.31 4.543 5.14 9.93 6.879 0.10
L 1.78 1.22 4.42 6.96 4.62 5.27 11.09 0.78 3.89 15.32 5.39 10.44 4.64 5.14 9.82 6.99 0.10
4
3
th
Little’s
Little’s Law Law
forfor Telesalescalls,
Telesales calls,October
October 10 , 2001:
10th, US US
2001: Bank
Bank
λ: Throughput Rate, Telesales, October 10th, 2001; US Bank
Arrivals to queue Telesales
10 October 2001
500
450
400
Number of cases 350
300
250
200
150
100
50
0
7:00 9:00 11:00 13:00 15:00 17:00 19:00 21:00 23:00
700
600
500
Means, Seconds
400
300
200
100
0
7:00 9:00 11:00 13:00 15:00 17:00 19:00 21:00 23:00
200
180
160
Average Number in Queue
140
120
100
80
60
40
20
0
7:00 9:00 11:00 13:00 15:00 17:00 19:00 21:00 23:00
time (resolution 30 min)
Time 7:00 7:30 8:00 8:30 9:00 9:30 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00
λ 76 102 182 262 379 464 440 433 410 431 422 418 401 439 453 432 373
W 109.8 123.8 383.5 403.7 503.5 522.5 607.9 602.1 552.4 521.1 508.6 468.8 442.1 467.3 545.9 483.1 442.1
λ*W 4.63 7.01 38.77 58.76 106.01 134.69 148.60 144.84 125.82 124.77 119.23 108.86 98.48 113.98 137.39 115.93 91.61
L 4.28 6.91 31.73 54.36 96.50 140.70 168.10 174.34 166.14 146.13 154.48 137.47 118.29 121.44 144.07 146.01 119.83
Time 15:30 16:00 16:30 17:00 17:30 18:00 18:30 19:00 19:30 20:00 20:30 21:00 21:30 22:00 22:30 23:00 23:30
λ 405 427 298 242 182 134 132 134 112 105 105 87 80 55 45 28 30
W 419.2 442.2 458.8 387.9 415.1 357.1 121.6 179.8 267.9 445.7 536.0 416.9 403.9 326.0 463.6 187.3 0.9
λ * W 94.31 104.89 75.96 52.15 41.97 26.58 8.92 13.38 16.67 26.00 31.27 20.15 17.95 9.96 11.59 2.91 0.02
L 107.86 101.22 111.60 82.93 42.23 32.32 10.57 13.24 18.67 21.07 32.50 24.10 20.33 10.69 11.13 4.35 0.02
5
4
rd
Little’s
Little’s LawLaw
forfor Russiancalls,
Russian calls,May
May 23 , 2005:
23rd, Israeli
2005: Telecom
Israeli Telecom
λ: Throughput Rate, Russian, May 23rd, 2005; Israeli Telecom
Arrivals to queue Russian
23 May 2005
90
80
70
Number of cases 60
50
40
30
20
10
0
7:00 9:00 11:00 13:00 15:00 17:00 19:00 21:00 23:00
250
200
Means, Seconds
150
100
50
0
7:00 9:00 11:00 13:00 15:00 17:00 19:00 21:00 23:00
Time (Resolution 30 min.)
10
8
Average Number in Queue
0
7:00 9:00 11:00 13:00 15:00 17:00 19:00 21:00 23:00
time (resolution 30 min)
Time 7:00 7:30 8:00 8:30 9:00 9:30 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00
λ 12 12 22 46 59 59 36 52 43 56 81 61 80 46 67 56 50
W 16.9 1.3 11.4 166.0 148.9 88.7 7.9 27.4 0.7 3.9 20.3 74.9 125.4 46.3 41.4 47.9 37.7
λ*W 0.11 0.01 0.14 4.24 4.88 2.91 0.16 0.79 0.02 0.12 0.91 2.54 5.57 1.18 1.54 1.49 1.05
L 0.08 0.04 0.14 3.97 4.88 3.18 0.16 0.79 0.02 0.12 0.88 2.57 5.32 1.44 1.36 1.67 1.00
Time 15:30 16:00 16:30 17:00 17:30 18:00 18:30 19:00 19:30 20:00 20:30 21:00 21:30 22:00 22:30 23:00 23:30
λ 57 52 62 70 75 79 68 68 60 55 55 56 56 43 27 14 6
W 65.7 22.4 79.2 156.6 118.6 139.3 143.6 150.2 179.2 151.3 209.5 219.5 224.7 88.4 33.7 107.0 0.7
λ*W 2.08 0.65 2.73 6.09 4.94 6.11 5.42 5.68 5.97 4.62 6.40 6.83 6.99 2.11 0.51 0.83 0.00
L 2.13 0.62 2.34 5.51 5.82 5.61 6.03 3.04 8.63 4.34 5.99 7.18 6.85 2.61 0.51 0.83 0.00
6
5
Motivation 1: λ customers/hour, each charged $1/hour while remaining in the system.
Then λ × W is the rate at which the system generates cash which, in turn, “clearly”
equals L.
Motivation 2: If there is always a single customer (L = 1) in the system, and every
customer remains in the system W hours on average (customers arrive one after the other),
then λ = 1/W is clear. When there are L in the system, on the average, λ = L/W is just
one leap of faith.
Hint at a stochastic version: think of i.i.d sojourn times and use the Strong Law of Large
Numbers.
Motivation 3 (finite horizon): Consider a system that operates in a finite horizon
(interval of time), and think of customers that arrive and leave (discrete units).
Interval length is T .
Note: Little’s Law will work if the system is empty at time 0, and empty at time T .
Motivation 4 (work in cycles): Consider a system that operates in cycles of equal
durations and has the same statistical behavior during each cycle.
Cycle length is T .
Note: Little’s Law will work if the system is at the same level (not necessarily 0) at the
beginning and at the end of the cycle, and if all the customers that are in the system at
the beginning of the cycle leave the system before the end of the cycle.
This happens, for instance, if there is a moment during the cycle when the system becomes
empty (see Example 10 on page 12, or Serfozo’s treatment on page 18).
7
W7
A(T)=N # customers
1
time
0
0 T
L(t)
1
time
0
0 T
L(0)=0 L(T)=0
Define RT
1
L= T 0 L(t)dt
PA(T )
1
W = A(T ) k=1 Wk =⇒ L = λW
A(T )
λ=
T
Examples
1. Management Strategy and Control: Only two out of the three λ, L, W deter-
mine a strategy; the third is implicitly determined.
Scenario: λ = demand (projected), W = goal (set), L = means of monitoring W .
2. Inventory Management
L = average inventory;
W = average time in inventory;
λ = average throughput rate.
3. Services Management
Scenario:
3.1 A restaurant processes on average 1500 customers per day (=15 hours). On
average, there are 50 customers waiting to place an order, waiting for an order to
arrive or eating.
9
λ = 1500 customers/day = 100 customers/hour;
L = 50 customers;
W = L/λ = 50/100 = 1/2 hours, average time in the restaurant.
5.1 Cars flow through a highway. We wish to relate the 3 quantities: Highway
Density, Flow Rate, Car Velocity.
System = 1 km of highway
L = avg. number of cars in system (1 km) = Density
λ = Flow, in avg. number of cars per hour (in = out = through)
W = avg. time to travel 1 km, say in hours
1
⇒ W
= Velocity, in km/hr; denote it V .
By Little’s Law:
Flow
Density =
Velocity
10
5.2 Cars flow over a single-loop detector, that can measure Occupancy = % time
there is a car above the detector;
Flow = avg. # cars per hour.
System = Detector
L = Occupancy (E [Indicator])
λ = Flow
`
W = V
time to traverse one detector
where V = Velocity, ` = av. car length.
By Little’s Law:
Flow × car-length
Occupancy = × 100%
Velocity
5.3 Empirically, transportation flow reveals the following “flow vs. occupancy”
relation (”flow vs. density” would look the same):
6:45 am
9:00 am
Speed = 60 mph
Depth of
Congestion
Critical
Occupancy Level
Figure 6: Flow vs. occupancy on a section at postmile 37.18 on I-10W, midnight to noon
on October 3, 2000.
11
From “The freeway congestion paradox”,
Chao Chen and Pravin Varaiya, 2001.
12:00
4:00 5:10
6:00
6:10 5:25
11:00 5:30
7:00
Figure 1 Congestion begins at 5:20 am. By 7:00 am, both speed and flow
have dropped dramatically.
Maximum
Flow
Depth of
Recovery Phase Congestion
60 mph
Occupancy (%)
Critical Occupancy
Level
a dynamic model that exhibits such transitions. Measurements of the recovery phase show
erratic fluctuations as in Fig. 6. The model in turn supports the following hypothesis about
traffic behavior:
If a metering policy keeps occupancy below its critical level in every section,
12 be maintained at 60 mph, and highway
efficiency will be 100 %, speed will
congestion will be prevented. A consequence of the metering is that vehicles
will be stopped at the ramps for some time.
The critical occupancy is the occupancy-level beyond which congestion starts
building up.
Note: For each point on the curve, the slope of the line connecting it with the
origin is proportional (equal) to the velocity; indeed:
This explains the (almost) straight line to the left of the critical occupancy: its slope
is the congestion-free velocity (60 miles/hr in California highways).
Note: with a single-loop detector covering N lanes, and assuming that traffic is
evenly divided among the lanes (though typically this is not the case), the Occu-
pancy should be calculated by using Flow/N , instead of merely Flow.
α · Pab = θ · Lq .
13
Thus, the company hires an average of 194.6 + 144 = 338.6 new employees per
month, or equivalently, 338.6 × 12 = 4063.2 new employees per year.
The 4063.2
data is from a bank call center. Each point corresponds to a 15-minute period of
0.1128 × 100%
a day (Sunday=to Thursday), = 11.28%
starting labor
at 7:00am, turnover
ending duringand
at midnight, a year
averaged
over36,
the000
whole year of 1999.
= 2.82% turnover during a 3-month period
• Why a positive y-intercept?
(compared with 40% at fast food, for example
• What about experienced customers?
and about 100% in many Call Centers).
Inventory/Queue
Inventory/Queue Build-up Diagram.
Build-up Diagram.
30
28
26
24
22
20
18
inventory L(t)
16
14
12
10
8
6
4
2
0
0 1 2 3 4 5 6 7 8 9 10 11 12
time (months)
4 4 2 2
17 × 12
+ 24 × 12
+ 12 × 12
+ 5× 12
=
17 5
17 × 4
3
12
+ + 24 ×8 12
4 + 12 ×
+ 2 2
12
++ 5 ×
6
2=
12
=16.5.
16.5
14
9. Shop Flow Control: JIT (Just-In-Time) principles advocate limiting the number
of active jobs (those that have been released to the shop floor).
Scenario: A job shop with Lold = 300 active jobs, Wold = 20 weeks, λold = 15
jobs/week.
Management familiar with Little’s law and JIT principles imposes Lhope ≤ 150 active
Lhope
jobs, in anticipation of λhope = 15 jobs/week, Whope = λhope ≤ 10 weeks.
It turns out, however, that
Lactual
Lactual ≤ 150, Wactual = 20 weeks, λactual = ≤ 7.5 jobs/week.
Wactual
What is lacking? Congestion curves (Strategic Q-theory): later.
L = average WIP;
W = average production time of a unit;
λ = average production rate.
15
11. Insurance: An insurance company processes 10,000 claims per year. The average
processing time of a claim is 3 weeks. Assuming 50 weeks per year, we have
12. Cash Flow (Accounts Receivable): A company sells 300M$ worth of finished
goods per year. The average amount of accounts receivable is 45M$.
λ = 300M$/year;
L = 45M$;
W = L/λ = 45/300 = 0.15 years = 1.8 months.
So it takes, on average, 1.8 months from the time a customer is billed until the time
payment is received.
13. Cash Flow: A paper mill processes 40M$ of raw material per year. Direct conver-
sion costs are 20M$ per year. Average inventory cost (raw material + conversion)
is 5M$.
Thus, there is an average lag of one month between the time a dollar enters the
system in the form of raw material (example: logs) or conversion cost (example:
chemicals), and the time it leaves the system in the form of finished goods (example:
paper).
14. Loss Queues: Customers arrive at a service facility at rate α. A fraction β of them
are blocked (do not enter). The others join a queue and wait until being served.
Assuming existence of averages and flow conservation, let
τ = average service time,
ν = long-run time-average number of customers in service. (Think G/G/S/N.)
Then
ν
β =1− · (Any three of (α, β, τ, ν) determine the fourth.)
ατ
By: system = servers, L = ν, W = τ, λ = α(1 − β).
16
15. Little’s Law in the “Production of Justice”.
8 0
01 . (6.2, 7.4) . (13.5, 7.4)
7 3 3
Avg. Months - W
01
6 3
3
(12, 4.9)
. 0
5 (7.2, 4.6) . 0 3
(26.3, 4.5)
01 0 .
4 01
01
3
2
1
0
0 5 10 15 20 25 30
Avg. Cases / Month - λ
8 0
01 . (6.2,
45 7.4) . (13.5,
100 7.4)
7 3 3
Avg. Months - W
01
6 3
59 0
5 3 .
(12, 4.9)
3
(7.2, 4.6) . 33 0 118
(26.3, 4.5)
.
01 0
4 01
01
3
2
1
0
0 5 10 15 20 25 30
Avg. Cases / Month - λ
17
Little’s Formula. Deterministic Model.
Infinite Horizon (Stidham’s formulation)
Averages: L=λ·W
Rigorous formulation
The system is characterized by {(An , Dn ), n ≥ 1}, where
An – time of the nth arrival.
Dn – departure-time of the nth arrival.
0 ≤ An ≤ An+1 An ≤ Dn < ∞ .
Define:
A(t) = number of arrivals until t;
D(t) = number of departures until t;
L(t) = number of units such that An ≤ t < Dn , i.e., number of units in the system;
Wn = Dn − An , sojourn time of the nth unit in the system.
Then
N
1 X 1ZT
lim Wn = W exists ⇔ lim L(t)dt = L exists,
N ↑∞ N T ↑∞ T 0
n=1
in which case L = λW .
18
Extension (Brumelle)
Associate with every n a corresponding function fn (t), t ∈ [An , Dn ].
Assume that fn (t) = 0, if t 6∈ [An , Dn ].
Interpret fn (t) as income-rate at time t (average income per unit of time).
Define
N Z ∞
1 X
GN = fn (t)dt (average income per customer);
N n=1 0
∞
1ZTX
HT = fn (t)dt (average income per time unit).
T 0 n=1
Then,
lim GN = G exists ⇔ lim HT = H exists,
N ↑∞ T ↑∞
in which case
H =λ·G
19
Stochastic example: M/M/1
Model
Birth-and-death process, birth rate λ, death rate µ.
Assumption
λ
ρ= µ
< 1, answers existence of stationary (limit) distribution π:
πk = (1 − ρ)ρk ,
k = 0, 1, 2, . . . (geometric distribution).
∞
X ρ λ
L = kπk = = .
k=0 1−ρ µ−λ
1 1 1 1
Little: W = L= = .
λ µ−λ µ 1−ρ
Check out:
∞
X
W = (PASTA) = E[sojourn time/k customers in system] πk
k=0
∞
" #
X k 1
= (memoryless property) = + πk
k=0 µ µ
1 1 1 1
= + L = ··· = .
µ µ µ 1−ρ
1 1 ρ
System = queue: Lq = λ Wq , Wq = W − µ
= µ 1−ρ
.
Lq – queue-length,
Wq – waiting-time.
System = server:
1
L=λ· ,
µ
L = ρ = probability that the system is not empty (customer waits)
= proportion of time when the server is busy (traffic intensity).
20
Stochastic Model (à la Serfozo1 )
{(An , Dn ), n ≥ 1} random variables; limits are a.s. (with probability 1)
1 1ZT
e.g. λ = lim A(t) a.s. ; L(t)dt → L a.s., as T ↑ ∞, etc.
t↑∞ t T 0
τn+1
1. τn ∼ τn+1 i.e. lim τn
= 1 a.s. (implied, for example, by τn /n → c).
n↑∞
2. For all n, there exists t ∈ [τn , τn+1 ) such that A(t) = D(t), i.e. L(t) = 0.
Theorem. If a system is periodically empty, the existence of any two positive limits out
of (L, λ, W ) implies existence of the third, as well as the relation L = λW .
Typical application: τn starts a “cycle” (eg. empty system; state 7), which gives rise
to a regenerative structure (eg. Markovian).
1
Introduction To Stochastic Networks, Springer 1999, Chapter 5
21
Application of H = λG : Brumelle’s formula (1971), in Whitt, pg. 257.
Let
Sn
An ≤ t < An + Wn
fn (t) = Sn + Wn + An − t An + Wn ≤ t ≤ An + Wn + Sn = Dn
0 otherwise.
N Z ∞ N
1 X 1 X 1 1
G = lim fn (t)dt = lim Sn Wn + Sn2 = E(SWq ) + E(S 2 )
N ↑∞ N n=1 0 N 1 2 2
V (t) – Work load process (under FIFO in G/G/1, it is equal to virtual waiting time).
22
h i
1
Brumelle’s Formula E(V ) = λ E(SWq ) + 2
E(S 2 )
If, as usually assumed in G/G/1, service times are independent of waiting times,
1
E(V ) = λ E(S) E(Wq ) + (S 2 ) .
2
If ASTA = arrivals see time averages, as in the case of Poisson arrivals, and if we have a
single-server queue with FIFO, then
d
V = Wq .
1
⇒ E(Wq ) = λ E(S)E(Wq ) + E(S 2 ) , which yields
2
1 1
E(Wq ) = · λE(S 2 )
1 − λE(S) 2
1 1
= λE(S 2 ) (using λE(S) = ρ)
1−ρ 2
ρ 1 E(S 2 )
=
1 − ρ 2 E(S)
2
ρ 1+CS
σ2
Khinchine Pollatcheck E(Wq ) = E(S) 1−ρ 2
C2 = E2
Hall, Formula (5.64) (M/G/1)
350
300
250
Average Wait, sec
200
150
100
50
0
0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00
Utilization (Hourly Avg.)
23