1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer...
-
Upload
scott-fisher -
Category
Documents
-
view
215 -
download
0
Transcript of 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer...
![Page 1: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/1.jpg)
1
Real-Time Queueing Network Theory
Presented by Akramul Azim
Department of Electrical and Computer EngineeringUniversity of Waterloo, Canada
John P. Lehoczky
![Page 2: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/2.jpg)
2
Outline
• Fundamentals
– Queueing theory– RT queueing theory
• Contributions
– RT network queueing theory– Analysis– Simulation studies– Reducing lateness
• Critics and Reviews
2
![Page 3: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/3.jpg)
3
Queueing Theory - Basics
• Queuing Theory deals with systems of the following
type:
InputProcess
OutputProcess
Queue
Example Input Process Output Process
Bank Customers arrive at
bank
Tellers serve the
customers
![Page 4: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/4.jpg)
4
Queueing Theory - Terminology and Notation
• State of the system
Number of customers in the queueing system (includes customers
in service)
• Queue length
Number of customers waiting for service
= State of the system - number of customers being served
• M/M/1 Queue
Single Server/Queue, arrivals follow poisson process, and job
service times have an exponential distribution
![Page 5: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/5.jpg)
5
Queueing Theory - Terminology and Notation
• N(t) = State of the system at time t, t ≥ 0
• Pn(t) = Probability that exactly n customers are in the queueing system at time t
• n = Mean arrival rate (expected # arrivals per unit time) of new customers when n customers are in the system
• s = Number of servers/queues (parallel service channels)
• n = Mean service rate for overall system (expected # customers completing service per unit
time) when n customers are in the system
![Page 6: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/6.jpg)
6
Real-Time Queueing Theory
• Applications/tasks/jobs/packets have explicit timing requirements (deadlines)
• Stochastic task arrivals, execution time, and network routing
• Avoid over-provisioning of resources
![Page 7: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/7.jpg)
7
Real-Time Queueing Network Theory
• Real-time network queueing theory combines,
1) Hard real-time system scheduling theory
2) Heavy traffic queueing theory
3) Jackson network theory
![Page 8: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/8.jpg)
8
Hard Real-time System Scheduling Theory
• Typical hard real-time system requirements
• Scheduling policies like EDF
![Page 9: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/9.jpg)
9
Heavy Traffic Queueing Theory
• Heavy traffic is the case when traffic intensity nears 1
• Timing behaviour of a RT queue becomes nearly deterministic in heavy traffic
![Page 10: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/10.jpg)
10
Jackson Network
• A Jackson network is connects queue (e.g., M/M/1) at each node
• Tasks may arrive any node in the network
• Tasks have end-to-end deadlines
![Page 11: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/11.jpg)
11
RT Queueing Theory Terminologies
• Lead-time (l): The time remaining until the task’s deadline.
• Lateness: Negative lead-times
• State variable: (N,l1,…,lN), where N is the number of tasks
• Lead time vector: (l1(t),…,lQ(t(t)), where Q(t) is the number in the system at time t
![Page 12: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/12.jpg)
12
RT Queueing Network Terminologies
• P = (pij): K ×K routing matrix where pjk represents the probability
that a task leaving node j goes to node k
• = (1,…, K), the arrival rates for independent Poisson external
task arrival process to each node
• Γ = (Г1 ,…, Гk), initial deadline distribution
• μ= (μ1, …, μk) , the service rates
• θ= (θ1, …, θk), the total arrival rate to each node
• ρj = θj/ μj , the traffic intensity of total task arrivals
![Page 13: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/13.jpg)
13
Facts – Queueing Theory
• Lead-time vector depends on three factors:
1. The number of tasks in the queue
2. The tasks deadline distribution
3. The queue discipline (scheduling policy)
![Page 14: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/14.jpg)
14
Facts – Queueing Network Theory
• Lead-time vector of a node depends on four factors:
1. The total task arrival rate
2. The tasks deadline distribution of that node
3. The queue discipline (scheduling policy)
4. the occupancy of the node (queue length)
![Page 15: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/15.jpg)
15
Experimental Analysis
• Simulation study performed
• Assume 2 stations in the network
• Nodes have a Queue, λ= .95, traffic intensity = 0.95
• Task deadlines are at least 40
• Two clocks are used: (1) Global (2) Local
• Global clock advances constantly
• Local clock advances only when queue lengths satisfy
some precondition
![Page 16: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/16.jpg)
16
Queue 1: Deadlines = 60 + 30 exp(1), Local Clk= 20
Note: 60 + 30 exp(1) = 141.54
![Page 17: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/17.jpg)
17
Queue 1: Deadlines = 60 + 30 exp(1), , Local Clk= 40
![Page 18: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/18.jpg)
18
Queue 1: Deadlines = 60 + 30 exp(1), , Local Clk= 60
![Page 19: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/19.jpg)
19
Queue 2: Deadlines = 60 + 30 exp(1), , Local Clk= 20
![Page 20: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/20.jpg)
20
Queue 2: Deadlines = 60 + 30 exp(1), , Local Clk= 40
![Page 21: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/21.jpg)
21
Queue 2: Deadlines = 60 + 30 exp(1), , Local Clk= 60
![Page 22: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/22.jpg)
22
Queue 1: Deadlines = 30 + 60 exp(1), , Local Clk= 20
Note: 30 + 30 exp(1) = 193.09
![Page 23: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/23.jpg)
23
Queue 1: Deadlines = 30 + 60 exp(1), , Local Clk= 40
![Page 24: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/24.jpg)
24
Queue 1: Deadlines = 30 + 60 exp(1), , Local Clk= 60
![Page 25: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/25.jpg)
25
Queue 1: Deadlines = 30 + 60 exp(1), , Local Clk= 80
![Page 26: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/26.jpg)
26
Queue 2: Deadlines = 30 + 60 exp(1), , Local Clk= 20
![Page 27: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/27.jpg)
27
Queue 2: Deadlines = 30 + 60 exp(1), , Local Clk= 40
![Page 28: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/28.jpg)
28
Queue 2: Deadlines = 30 + 60 exp(1), , Local Clk= 60
![Page 29: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/29.jpg)
29
Queue 2: Deadlines = 30 + 60 exp(1), , Local Clk= 80
![Page 30: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/30.jpg)
30
Simulation Findings
• Queueing theory can analyze complex network
structures
• Task lead time profiles with reasonable accuracy at high,
but not extreme
![Page 31: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/31.jpg)
31
Facts - Lateness
• If task deadlines are 40, approximately 40% of the tasks
will miss their deadlines
• If task deadlines are increased to 60, still 20% will be
late.
• This holds even though the best scheduling policy used
![Page 32: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/32.jpg)
32
Control Policies to Reduce Lateness
• Blocking policy:
1. Tasks are blocked if the total number of tasks
reaches some level
2. Individual queue reaches some threshold
• Abort policy:
1. Tasks should be aborted as soon as they become late
![Page 33: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/33.jpg)
33
Blocking Probabilities (Mean task deadlines = 40)
Blocking strategy can reduce the task lateness from 40% to 4%
![Page 34: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/34.jpg)
34
Critics
• Unable to provide guarantee of meeting deadlines of all tasks in the worst-case
• Simulation study performed with distributions but not any real data set or application
• Worst-case resource constraints not taken into account
• Analysis on bounding buffer requirements is absent
![Page 35: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/35.jpg)
35
Conclusions and Future Work
• Can be used for modelling probabilistic real-time systems
• Can be used to assess queue control policies
• Can analyze large-scale soft-RT systems/networks
• Future work may include periodic arrival process and deriving
bounds on queue length or buffer requirements.
![Page 36: 1 Real-Time Queueing Network Theory Presented by Akramul Azim Department of Electrical and Computer Engineering University of Waterloo, Canada John P.](https://reader035.fdocuments.in/reader035/viewer/2022062421/56649ddd5503460f94ad646c/html5/thumbnails/36.jpg)
36
Thank You. Any Questions?
36