1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a...
Transcript of 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a...
![Page 1: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/1.jpg)
FLUID MODELS OF MANY-SERVER QUEUES
WITH ABANDONMENT
Jiheng Zhang
June 10, 2010
![Page 2: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/2.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Model and Motivation
Many-server queue
buffer
N
Motivation:
Customer call centers and other services areas.
Many-server Queues with Abandonment / Jiheng Zhang
![Page 3: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/3.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Model and Motivation
Many-server queue
buffer
N
Motivation:
Customer call centers and other services areas.
Many-server Queues with Abandonment / Jiheng Zhang
![Page 4: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/4.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Model and Motivation
Many-server queue with abandonment
buffer
N
Motivation:
Customer call centers and other services areas.
Many-server Queues with Abandonment / Jiheng Zhang
![Page 5: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/5.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Many-server queue v.s. Single-server queue
Large scale: high demand, need for high capacitySingle server queue: increase speed
Many-server queue: increase number of servers
Many-server Queues with Abandonment / Jiheng Zhang
![Page 6: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/6.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Many-server queue v.s. Single-server queue
Large scale: high demand, need for high capacitySingle server queue: increase speed
Many-server queue: increase number of servers
Many-server Queues with Abandonment / Jiheng Zhang
![Page 7: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/7.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
A Real World Challenge
The service time is not exponentially distributed!
Brown et. al. Statistical analysis of a telephone call center:a queueing-science perspective. JASA 2005
In this research
Arrival process: general
Service/patient time distribution: general
Many-server Queues with Abandonment / Jiheng Zhang
![Page 8: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/8.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Literature Review
Many-server Queues
Halfin and Whitt 1981 (M/M/N)
Puhalskii and Reiman 2000 (G/Ph/N)
Jelenkovic, Mandelbaum and Momcilovic 2004 (G/D/N)
Whitt 2005 (G/H∗2/n/m)
Garmarnik and Momcilovic 2007 (G/La/N)
Reed 2007, Puhalskii and Reed 2008 (G/G/N)
Mandelbaum and Momcilovic 2008 (G/G/N)
Kaspi and Ramanan 2009, Kaspi 2009 (G/G/N)
. . . . . .
Many-server Queues with Abandonment / Jiheng Zhang
![Page 9: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/9.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Literature Review
Many-server Queues with Abandonment
Whitt 2004 (M/M/N + M)
Zeltyn and Mandelbaum 2005 (G/M/N + G)
Whitt 2006 (G/G/N + G)
Puhalskii 2008 (Mt/Mt/Nt + Mt)
Kang and Ramanan 2008 (G/G/N + G)
Mandelbaum and Momcilovic 2009 (G/G/N + G)
Dai, He and Tezcan 2009 (G/Ph/N + G)
. . . . . .
Many-server Queues with Abandonment / Jiheng Zhang
![Page 10: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/10.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
System Dynamics – example with N = 2
3 t = 0
25 arrival t = 1
2 14 arrival t = 2
23 departure t = 3
12 t = 4
1 departure t = 5
Many-server Queues with Abandonment / Jiheng Zhang
![Page 11: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/11.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
System Dynamics – example with N = 2
3 t = 0
25 arrival t = 1
2 14 arrival t = 2
23 departure t = 3
12 t = 4
1 departure t = 5
Many-server Queues with Abandonment / Jiheng Zhang
![Page 12: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/12.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
System Dynamics – example with N = 2
3 t = 0
25 arrival t = 1
2 14 arrival t = 2
23 departure t = 3
12 t = 4
1 departure t = 5
Many-server Queues with Abandonment / Jiheng Zhang
![Page 13: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/13.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
System Dynamics – example with N = 2
3 t = 0
25 arrival t = 1
2 14 arrival t = 2
23 departure t = 3
12 t = 4
1 departure t = 5
Many-server Queues with Abandonment / Jiheng Zhang
![Page 14: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/14.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
System Dynamics – example with N = 2
3 t = 0
25 arrival t = 1
2 14 arrival t = 2
23 departure t = 3
12 t = 4
1 departure t = 5
Many-server Queues with Abandonment / Jiheng Zhang
![Page 15: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/15.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
System Dynamics – example with N = 2
3 t = 0
25 arrival t = 1
2 14 arrival t = 2
23 departure t = 3
12 t = 4
1 departure t = 5
Many-server Queues with Abandonment / Jiheng Zhang
![Page 16: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/16.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Measure-valued State Descriptor
Server pool
Z(t)(C): # of customers in server with remaining servicetime in C ⊂ (0,∞)
Evolution
Z(t0 + t)(C) = Z(t0)(C + t) + . . . (t0, t0 + t) . . .
Richness
Z(t) = Z(t)((0,∞))
LiteratureGromoll, Puha & Williams ’02, Puha &Williams ’02, Gromoll ’06
Gromoll & Kurk ’07, Gromoll, Robert & Zwart ’08, . . .
Zhang, Dai & Zwart ’07, ’08, Zhang & Zwart ’08
Many-server Queues with Abandonment / Jiheng Zhang
![Page 17: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/17.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Measure-valued State Descriptor
Server pool
Z(t)(C): # of customers in server with remaining servicetime in C ⊂ (0,∞)
Evolution
Z(t0 + t)(C) = Z(t0)(C + t) + . . . (t0, t0 + t) . . .
Richness
Z(t) = Z(t)((0,∞))
LiteratureGromoll, Puha & Williams ’02, Puha &Williams ’02, Gromoll ’06
Gromoll & Kurk ’07, Gromoll, Robert & Zwart ’08, . . .
Zhang, Dai & Zwart ’07, ’08, Zhang & Zwart ’08
Many-server Queues with Abandonment / Jiheng Zhang
![Page 18: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/18.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Measure-valued State Descriptor
Server pool
Z(t)(C): # of customers in server with remaining servicetime in C ⊂ (0,∞)
Evolution
Z(t0 + t)(C) = Z(t0)(C + t) + . . . (t0, t0 + t) . . .
Richness
Z(t) = Z(t)((0,∞))
LiteratureGromoll, Puha & Williams ’02, Puha &Williams ’02, Gromoll ’06
Gromoll & Kurk ’07, Gromoll, Robert & Zwart ’08, . . .
Zhang, Dai & Zwart ’07, ’08, Zhang & Zwart ’08
Many-server Queues with Abandonment / Jiheng Zhang
![Page 19: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/19.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Measure-valued State Descriptor
Server pool
Z(t)(C): # of customers in server with remaining servicetime in C ⊂ (0,∞)
Evolution
Z(t0 + t)(C) = Z(t0)(C + t) + . . . (t0, t0 + t) . . .
Richness
Z(t) = Z(t)((0,∞))
LiteratureGromoll, Puha & Williams ’02, Puha &Williams ’02, Gromoll ’06
Gromoll & Kurk ’07, Gromoll, Robert & Zwart ’08, . . .
Zhang, Dai & Zwart ’07, ’08, Zhang & Zwart ’08
Many-server Queues with Abandonment / Jiheng Zhang
![Page 20: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/20.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Measure-valued State Descriptor
Virtual buffer
R(t)(C): # of customers in virtual buffer with remainingpatient time in C ⊂(−∞,∞)
Evolution
buffer
Richness
Q(t) = R(t)((0,∞))
R(t) = R(t)((−∞,∞))
Many-server Queues with Abandonment / Jiheng Zhang
![Page 21: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/21.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Measure-valued State Descriptor
Virtual buffer
R(t)(C): # of customers in virtual buffer with remainingpatient time in C ⊂(−∞,∞)
Evolution
buffer
Richness
Q(t) = R(t)((0,∞))
R(t) = R(t)((−∞,∞))
Many-server Queues with Abandonment / Jiheng Zhang
![Page 22: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/22.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Measure-valued State Descriptor
Virtual buffer
R(t)(C): # of customers in virtual buffer with remainingpatient time in C ⊂(−∞,∞)
Evolution
buffer
61214
Richness
Q(t) = R(t)((0,∞))
R(t) = R(t)((−∞,∞))
Many-server Queues with Abandonment / Jiheng Zhang
![Page 23: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/23.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Measure-valued State Descriptor
Virtual buffer
R(t)(C): # of customers in virtual buffer with remainingpatient time in C ⊂(−∞,∞)
Evolution
buffer
50103
Richness
Q(t) = R(t)((0,∞))
R(t) = R(t)((−∞,∞))
Many-server Queues with Abandonment / Jiheng Zhang
![Page 24: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/24.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Measure-valued State Descriptor
Virtual buffer
R(t)(C): # of customers in virtual buffer with remainingpatient time in C ⊂(−∞,∞)
Evolution
buffer
4-10-12
Richness
Q(t) = R(t)((0,∞))
R(t) = R(t)((−∞,∞))
Many-server Queues with Abandonment / Jiheng Zhang
![Page 25: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/25.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Measure-valued State Descriptor
Virtual buffer
R(t)(C): # of customers in virtual buffer with remainingpatient time in C ⊂(−∞,∞)
Evolution
buffer
4-10-12
Richness
Q(t) = R(t)((0,∞))
R(t) = R(t)((−∞,∞))
Many-server Queues with Abandonment / Jiheng Zhang
![Page 26: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/26.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
System Dynamics
Internal transfer process
B(t) = E(t)− R(t)
1 + B(t): index of the next customer to be served
Stochastic dynamic equations
δui−(t−ai)C
δvi−(t−τi)C
R(t)(C) =
E(t)∑i=1+B(t)
C ∈ B(R)
Z(t)(C) = Z(0)(C + t)
+
B(t)∑i=1+B(0)
1{ui>τi−ai}C ∈ B(R+)
Many-server Queues with Abandonment / Jiheng Zhang
![Page 27: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/27.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
System Dynamics
Internal transfer process
B(t) = E(t)− R(t)
1 + B(t): index of the next customer to be served
Stochastic dynamic equations
δui−(t−ai)C
δvi−(t−τi)C
R(t)(C) =
E(t)∑i=1+B(t)
C ∈ B(R)
Z(t)(C) = Z(0)(C + t)
+
B(t)∑i=1+B(0)
1{ui>τi−ai}C ∈ B(R+)
Many-server Queues with Abandonment / Jiheng Zhang
![Page 28: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/28.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
System Dynamics
Internal transfer process
B(t) = E(t)− R(t)
1 + B(t): index of the next customer to be served
Stochastic dynamic equations
δui−(t−ai)C
δvi−(t−τi)C
R(t)(C) =
E(t)∑i=1+B(t)
δui(C + t − ai), C ∈ B(R)
Z(t)(C) = Z(0)(C + t)
+
B(t)∑i=1+B(0)
1{ui>τi−ai} δvi(C + t − τi),C ∈ B(R+)
Many-server Queues with Abandonment / Jiheng Zhang
![Page 29: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/29.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
System Dynamics
Internal transfer process
B(t) = E(t)− R(t)
1 + B(t): index of the next customer to be served
Stochastic dynamic equationsδui−(t−ai)C
δvi−(t−τi)C
R(t)(C) =
E(t)∑i=1+B(t)
δui(C + t − ai), C ∈ B(R)
Z(t)(C) = Z(0)(C + t)
+
B(t)∑i=1+B(0)
1{ui>τi−ai} δvi(C + t − τi),C ∈ B(R+)
Many-server Queues with Abandonment / Jiheng Zhang
![Page 30: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/30.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
System Dynamics
Internal transfer process
B(t) = E(t)− R(t)
1 + B(t): index of the next customer to be served
Stochastic dynamic equationsδui−(t−ai)C
δvi−(t−τi)C
R(t)(C) =
E(t)∑i=1+B(t)
δui(C + t − ai), C ∈ B(R)
Z(t)(C) = Z(0)(C + t)
+
B(t)∑i=1+B(0)
1{ui>τi−ai} δvi(C + t − τi),C ∈ B(R+)
Many-server Queues with Abandonment / Jiheng Zhang
![Page 31: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/31.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
System Dynamics
Internal transfer process
B(t) = E(t)− R(t)
1 + B(t): index of the next customer to be served
Stochastic dynamic equationsδui−(t−ai)C
δvi−(t−τi)C
R(t)(C) =
E(t)∑i=1+B(t)
δui(C + t − ai), C ∈ B(R)
Z(t)(C) = Z(0)(C + t)
+
B(t)∑i=1+B(0)
1{ui>τi−ai} δvi(C + t − τi),C ∈ B(R+)
Many-server Queues with Abandonment / Jiheng Zhang
![Page 32: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/32.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
System Dynamics
Total number of customers
X(t) = Q(t) + Z(t)
Policy constraints
Q(t) = (X(t)− N)+, Z(t) = (X(t) ∧ N)
Many-server Queues with Abandonment / Jiheng Zhang
![Page 33: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/33.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Fluid Model
E(·): λ·{ui}: F (ϑF ∼ F)
{vi}: G (ϑG ∼ G)
B(s) = λs− R(s)
Fluid dynamic equations
R(t)(C) =
∫ t
t− R(t)λ
ϑF(C + t − s)dλs, C ∈ B(R)
Z(t)(C) = Z(0)(C + t)
+
∫ t
0ϑF(
R(s)λ,∞)ϑG(C + t − s)dB(s),
C ∈ B(R+)
Constraints
Q(t) = (X(t)− N)+, Z(t) = (X(t) ∧ N)
Many-server Queues with Abandonment / Jiheng Zhang
![Page 34: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/34.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Fluid Model
E(·): λ·{ui}: F (ϑF ∼ F)
{vi}: G (ϑG ∼ G)
B(s) = λs− R(s)
Fluid dynamic equations
R(t)(C) =
∫ t
t− R(t)λ
ϑF(C + t − s)dλs, C ∈ B(R)
Z(t)(C) = Z(0)(C + t)
+
∫ t
0ϑF(
R(s)λ,∞)ϑG(C + t − s)dB(s),
C ∈ B(R+)
Constraints
Q(t) = (X(t)− N)+, Z(t) = (X(t) ∧ N)
Many-server Queues with Abandonment / Jiheng Zhang
![Page 35: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/35.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Fluid Model
E(·): λ·{ui}: F (ϑF ∼ F)
{vi}: G (ϑG ∼ G)
B(s) = λs− R(s)
Fluid dynamic equations
R(t)(C) =
∫ t
t− R(t)λ
ϑF(C + t − s)dλs, C ∈ B(R)
Z(t)(C) = Z(0)(C + t)
+
∫ t
0ϑF(
R(s)λ,∞)ϑG(C + t − s)dB(s),
C ∈ B(R+)
Constraints
Q(t) = (X(t)− N)+, Z(t) = (X(t) ∧ N)
Many-server Queues with Abandonment / Jiheng Zhang
![Page 36: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/36.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Fluid Model
E(·): λ·{ui}: F (ϑF ∼ F)
{vi}: G (ϑG ∼ G)
B(s) = λs− R(s)
has to be increasing!
Fluid dynamic equations
R(t)(C) =
∫ t
t− R(t)λ
ϑF(C + t − s)dλs, C ∈ B(R)
Z(t)(C) = Z(0)(C + t)
+
∫ t
0ϑF(
R(s)λ,∞)ϑG(C + t − s)dB(s),
C ∈ B(R+)
Constraints
Q(t) = (X(t)− N)+, Z(t) = (X(t) ∧ N)
Many-server Queues with Abandonment / Jiheng Zhang
![Page 37: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/37.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Fluid Model
E(·): λ·{ui}: F (ϑF ∼ F)
{vi}: G (ϑG ∼ G)
B(s) = λs− R(s)
has to be increasing!
Fluid dynamic equations
R(t)(C) =
∫ t
t− R(t)λ
ϑF(C + t − s)dλs, C ∈ B(R)
Z(t)(C) = Z(0)(C + t)
+
∫ t
0ϑF(
R(s)λ,∞)ϑG(C + t − s)dB(s),
C ∈ B(R+)
Constraints
Q(t) = (X(t)− N)+, Z(t) = (X(t) ∧ N)
Many-server Queues with Abandonment / Jiheng Zhang
![Page 38: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/38.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Existence and Uniqueness of Fluid Model Solution
Fluid model solution with initial condition (R0, Z0)
(R(0), Z(0)) = (R0, Z0)
(R(·), Z(·)) satisfies fluid dynamic equations and constraints
THEOREM
Assume that
G is continuous, with 0 < µ <∞,F(·) is Liptachitz continuous, or sup
x∈[0,∞)
hF(x) <∞.
There exists a unique solution to the fluid model (λ,F,G,N) forany valid initial condition (R0, Z0).
Many-server Queues with Abandonment / Jiheng Zhang
![Page 39: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/39.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Existence and Uniqueness of Fluid Model Solution
Fluid model solution with initial condition (R0, Z0)
(R(0), Z(0)) = (R0, Z0)
(R(·), Z(·)) satisfies fluid dynamic equations and constraints
THEOREM
Assume that
G is continuous, with 0 < µ <∞,F(·) is Liptachitz continuous, or sup
x∈[0,∞)
hF(x) <∞.
There exists a unique solution to the fluid model (λ,F,G,N) forany valid initial condition (R0, Z0).
Many-server Queues with Abandonment / Jiheng Zhang
![Page 40: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/40.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Invariant State of Fluid Model
Invariant state (R∞, Z∞)
(R(0), Z(0)) = (R∞, Z∞) implies (R(·), Z(·)) ≡ (R∞, Z∞)
THEOREM
The state (R∞, Z∞) is an invariant state if and only if it satisfies
R∞(Cx) = λ
∫ w
0Fc(x + s)ds, x ∈ R,
Z∞(Cx) = min (ρ, 1) N[1− Ge(x)], x ∈ R+,
where w is a solution to the equation
F(w) = max(ρ− 1ρ
, 0).
Many-server Queues with Abandonment / Jiheng Zhang
![Page 41: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/41.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Invariant State of Fluid Model
Invariant state (R∞, Z∞)
(R(0), Z(0)) = (R∞, Z∞) implies (R(·), Z(·)) ≡ (R∞, Z∞)
THEOREM
The state (R∞, Z∞) is an invariant state if and only if it satisfies
R∞(Cx) = λ
∫ w
0Fc(x + s)ds, x ∈ R,
Z∞(Cx) = min (ρ, 1) N[1− Ge(x)], x ∈ R+,
where w is a solution to the equation
F(w) = max(ρ− 1ρ
, 0).
Many-server Queues with Abandonment / Jiheng Zhang
![Page 42: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/42.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Fluid Model Analysis
Replace C by Cx = (x,∞), (note that ϑF(Cx) = Fc(x))
R(t)(Cx) = λ
∫ t
t− R(t)λ
Fc(x + t − s)ds, x ∈ R,
Z(t)(Cx) = Z(0)(Cx + t) +
∫ t
0Fc(
R(s)λ
)Gc(x + t − s)dB(s), x ∈ R+,
The functional equation
X(t) = ζ0(t) + ρ
∫ t
0H((X(t − s)− 1)+
)dGe(s) +
∫ t
0(X(t − s)− 1)+dG(s).
where H(x) = Fc(F−1e (
α
λx)).
Many-server Queues with Abandonment / Jiheng Zhang
![Page 43: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/43.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Fluid Model Analysis
Replace C by Cx = (x,∞), (note that ϑF(Cx) = Fc(x))
R(t)(Cx) = λ
∫ t
t− R(t)λ
Fc(x + t − s)ds, x ∈ R,
Z(t)(Cx) = Z(0)(Cx + t) +
∫ t
0Fc(
R(s)λ
)Gc(x + t − s)dB(s), x ∈ R+,
The functional equation
X(t) = ζ0(t) + ρ
∫ t
0H((X(t − s)− 1)+
)dGe(s) +
∫ t
0(X(t − s)− 1)+dG(s).
where H(x) = Fc(F−1e (
α
λx)).
Many-server Queues with Abandonment / Jiheng Zhang
![Page 44: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/44.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
The Special Case with Exponential Distribution
Now, we specialize in the case with exponential distribution, i.e.
F(t) = Fe(t) = 1− e−αt, G(t) = Ge(t) = 1− e−µt.
Now the key equation becomes
X(t) = ζ0(t) + ρ
∫ t
0
[1− α
λ
((X(t − s)− 1)+
)]µe−µsds
+
∫ t
0(X(t − s)− 1)+µe−µsds,
with ζ0(t) = X0e−µt. After some algebra, we get
X′(t) = µ(ρ− 1)− α(X(t)− 1)+ + µ(X(t)− 1)−. (Whitt 04)
Many-server Queues with Abandonment / Jiheng Zhang
![Page 45: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/45.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
The Special Case with Exponential Distribution
Now, we specialize in the case with exponential distribution, i.e.
F(t) = Fe(t) = 1− e−αt, G(t) = Ge(t) = 1− e−µt.
Now the key equation becomes
X(t) = ζ0(t) + ρ
∫ t
0
[1− α
λ
((X(t − s)− 1)+
)]µe−µsds
+
∫ t
0(X(t − s)− 1)+µe−µsds,
with ζ0(t) = X0e−µt.
After some algebra, we get
X′(t) = µ(ρ− 1)− α(X(t)− 1)+ + µ(X(t)− 1)−. (Whitt 04)
Many-server Queues with Abandonment / Jiheng Zhang
![Page 46: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/46.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
The Special Case with Exponential Distribution
Now, we specialize in the case with exponential distribution, i.e.
F(t) = Fe(t) = 1− e−αt, G(t) = Ge(t) = 1− e−µt.
Now the key equation becomes
X(t) = ζ0(t) + ρ
∫ t
0
[1− α
λ
((X(t − s)− 1)+
)]µe−µsds
+
∫ t
0(X(t − s)− 1)+µe−µsds,
with ζ0(t) = X0e−µt. After some algebra, we get
X′(t) = µ(ρ− 1)− α(X(t)− 1)+ + µ(X(t)− 1)−. (Whitt 04)
Many-server Queues with Abandonment / Jiheng Zhang
![Page 47: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/47.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Fluid Scaling and Limiting Regimes
A sequence of systems indexed by the number of servers n.
Fluid scaling
Rn(t) =1nRn(t), Zn(t) =
1nZn(t),
Arrival rate of the nth system λn ∼ nλ.
ρn =λn
nµn→ ρ ∈ (0,∞)
> 1, ED
= 1, QED
< 1, QD
Constraints
Qn(t) = (Xn(t)− 1)+, Zn(t) = (Xn(t) ∧ 1)
Many-server Queues with Abandonment / Jiheng Zhang
![Page 48: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/48.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Fluid Scaling and Limiting Regimes
A sequence of systems indexed by the number of servers n.
Fluid scaling
Rn(t) =1nRn(t), Zn(t) =
1nZn(t),
Arrival rate of the nth system λn ∼ nλ.
ρn =λn
nµn→ ρ ∈ (0,∞)
> 1, ED
= 1, QED
< 1, QD
Constraints
Qn(t) = (Xn(t)− 1)+, Zn(t) = (Xn(t) ∧ 1)
Many-server Queues with Abandonment / Jiheng Zhang
![Page 49: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/49.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Fluid Scaling and Limiting Regimes
A sequence of systems indexed by the number of servers n.
Fluid scaling
Rn(t) =1nRn(t), Zn(t) =
1nZn(t),
Arrival rate of the nth system λn ∼ nλ.
ρn =λn
nµn→ ρ ∈ (0,∞)
> 1, ED
= 1, QED
< 1, QD
Constraints
Qn(t) = (Xn(t)− 1)+, Zn(t) = (Xn(t) ∧ 1)
Many-server Queues with Abandonment / Jiheng Zhang
![Page 50: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/50.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Fluid Scaling and Limiting Regimes
A sequence of systems indexed by the number of servers n.
Fluid scaling
Rn(t) =1nRn(t), Zn(t) =
1nZn(t),
Arrival rate of the nth system λn ∼ nλ.
ρn =λn
nµn→ ρ ∈ (0,∞)
> 1, ED
= 1, QED
< 1, QD
Constraints
Qn(t) = (Xn(t)− 1)+, Zn(t) = (Xn(t) ∧ 1)
Many-server Queues with Abandonment / Jiheng Zhang
![Page 51: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/51.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Functional Law of Large Numbers
Assumption A:
1 En(·)⇒ λ·2 ϑn
F → ϑF, ϑnG → ϑG
3 µn → µ
4 (Rn(0), Zn(0))⇒ (R0, Z0)
5 R0 and Z0 has no atoms
THEOREM
Under assumption A
(Rn(·), Zn(·))⇒ (R(·), Z(·)) as n→∞,
where (R(·), Z(·)) is almost surely the fluid model solution to(λ,F,G, 1) with initial condition (R0, Z0).
Many-server Queues with Abandonment / Jiheng Zhang
![Page 52: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/52.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Performance Evaluation
Approximation formulas
E(W|S) = w, F(w) = max ((ρ− 1)/ρ, 0)
E(Q) =λ
αFe(w)
M/GI/100-GI, λ = 120, µ = 1, α = 1 (Whitt 2006)
Abd. Ser. E[Q] E[W|S]
E2E2 40.25± 0.057 0.353± 0.00051
LN(1, 4) 39.56± 0.097 0.343± 0.00094Approximation 41.11 0.365
LN(1, 4)E2 14.51± 0.018 0.126± 0.00017
LN(1, 4) 14.52± 0.043 0.125± 0.00027Approximation 14.63 0.131
Many-server Queues with Abandonment / Jiheng Zhang
![Page 53: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/53.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Performance Evaluation
Approximation formulas
E(W|S) = w, F(w) = max ((ρ− 1)/ρ, 0)
E(Q) =λ
αFe(w)
M/GI/100-GI, λ = 120, µ = 1, α = 1 (Whitt 2006)
Abd. Ser. E[Q] E[W|S]
E2E2 40.25± 0.057 0.353± 0.00051
LN(1, 4) 39.56± 0.097 0.343± 0.00094Approximation 41.11 0.365
LN(1, 4)E2 14.51± 0.018 0.126± 0.00017
LN(1, 4) 14.52± 0.043 0.125± 0.00027Approximation 14.63 0.131
Many-server Queues with Abandonment / Jiheng Zhang
![Page 54: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/54.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
A Missing Gap
Interchange of Steady State and Heavy Traffic Limits
(Rn(t), Zn(t)) (Rn∞, Zn
∞)t→∞
(R(t), Z(t))
n→∞
(R∞, Z∞)t→∞
n→∞
Many-server Queues with Abandonment / Jiheng Zhang
![Page 55: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/55.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
A Missing Gap
Interchange of Steady State and Heavy Traffic Limits
(Rn(t), Zn(t)) (Rn∞, Zn
∞)t→∞
(R(t), Z(t))
n→∞
(R∞, Z∞)t→∞
n→∞
Many-server Queues with Abandonment / Jiheng Zhang
![Page 56: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/56.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
A Missing Gap
Interchange of Steady State and Heavy Traffic Limits
(Rn(t), Zn(t)) (Rn∞, Zn
∞)t→∞
(R(t), Z(t))
n→∞
(R∞, Z∞)t→∞
n→∞
Many-server Queues with Abandonment / Jiheng Zhang
![Page 57: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/57.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
A Missing Gap
Interchange of Steady State and Heavy Traffic Limits
(Rn(t), Zn(t)) (Rn∞, Zn
∞)t→∞
(R(t), Z(t))
n→∞
(R∞, Z∞)t→∞
n→∞
Many-server Queues with Abandonment / Jiheng Zhang
![Page 58: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/58.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Questions?
Many-server Queues with Abandonment / Jiheng Zhang
![Page 59: 1pt Fluid Models of Many-server Queues with AbandonmentBrown et. al. Statistical analysis of a telephone call center: a queueing-science perspective. JASA 2005 In this research Arrival](https://reader034.fdocuments.in/reader034/viewer/2022051920/600d3bb699616b1d944c2d96/html5/thumbnails/59.jpg)
Background Stochastic Model Fluid Model Functional LLN Approximations
Thank you!
Many-server Queues with Abandonment / Jiheng Zhang