Download presentation
Presentation is loading. Please wait.
Published byPaulina Stone Modified over 9 years ago
1
Review Session Jehan-François Pâris
2
Agenda Statistical Analysis of Outputs Operational Analysis Case Studies Linear Regression
3
How to use this presentation Most problems have One slide stating the problem One slide explaining how to solve the problem One slide allowing you to check your answer You will learn more by trying first to do the problems on your own than by reading their solutions Do not forget either to review the problems in the original notes
4
Statistical Analysis of Outputs
5
The big picture The problems Constructing confidence intervals Handling auto correlated data The tools Central-Limit Theorem Wilson’s formula Batch means (and regeneration) RNG tricks
6
Confidence Intervals Distinguish between CIs for means CSIM does it for you CIs for proportions We are on our own Major issue is independence of data points CSIM uses batch means
7
Central Limit Theorem If the n mutually independent random variables x 1, x 2, …, x n have the same distribution, and if their mean and their variance 2 exist then …
8
Central Limit Theorem The random variable is distributed according to the standard normal distribution (zero mean and unit variance).
9
CI for means (I) For large values of n, the (1- )% confidence interval for is given by with
10
CI for means (II) F(z) is taken from a table of the normal distribution F(0.025) = 1.96 For smaller values of n, we have to use Student’s t random variable Wider CIs We replace by the sample standard deviation s
11
Example We have 100 observations for the waiting time xbar = 4.25 minutes s 2 = 25
12
Example We have 100 observations for the waiting time xbar = 4.25 minutes s 2 = 25 Answer is 4.25 ± 1.96 sqrt(25/100) = 4.25 ± 0.98
13
CI for proportions A proportion represents the probability P(X ) for some fixed threshold 97% of our customers have to wait less than one minute Distributed according to a binomial law Use Wilson ’ s formula
14
Wilson’s formula When n > 29, we can use the Wilson’s interval where z /2 = 1.96 for a 95% C.I.
15
Example We have want to estimate the proportion of packets that wait more than four slots 400 observations 40 packets waited more than four slots
16
Answer Divisor: 1 + 1.96 2 /400 1.01 (instead of 1.0096) Central term 0.1 + 1.96 2 /(2×400) 0.105 (instead of 1.048) Half width sqrt( (0.1×0.9)/400 + 1.96 2 /(2×400 2 ) ) sqrt (0.09/400 + (4/800)/400) 1/20 sqrt (0.09 +0.0025) 0.3/20 = 0.015 Result is (0.105 ± 0.015)/ 1.01 = 0.104 ± 0.015
17
Batch means (I) Simulation data are often autocorrelated Packet delays in ALOHA Waiting times in queues … Batch means reduce (but do not completely eliminate) that effect
18
Batch means (II) Group measurements into fixed-size batches of consecutive data Compute mean of each batch If batches are large enough, these means will be independent Can use standard-limit theorem, … In case of doubt, compute autocorrelation function for successive batch means
19
Regeneration (I) The idea Partition simulation data into intervals such that Data measured inside the same interval might be correlated Data measured in different intervals are independent
20
Regeneration (II) How? System goes to a regeneration point each time Its queues become empty All the disk drives are operational … Criterion is system specific
21
Streams When you want to evaluate two different configurations of a system, it is often good idea to use separate random number streams for arrivals and service times Arrival times remain unchanged when we change other parameters of the system
22
Operational Analysis
23
Single server (I) We can measure T the length of the observation period A the number of arrivals during the observation period B the total amount of busy times during the observation period C the number of completions during the observation period
24
Single server (II) We can compute = A/Tthe arrival rate X = C/Tthe output rate U = B/Tthe utilization S = B/Cthe mean service time There are two ways to compute U U = B/T = (C/T )(B/C) = XS In general A C and X
25
Little’s law If W is the total time spent by all tasks inside the system over the observation period, then N = W/T R = W/C Since W/T = (C/T)(W/C) = XR, N = XR This is important
26
A problem An ice-cream parlor Observed during 6 hours Visited by 120 customers Spend an average of 24 minutes inside What is the average number of customers inside the parlor?
27
Answer We compute X and apply Little’s Law
28
Answer We compute X and apply Little’s Law X = 120/6 = 20 customers/hour R = 24 minutes = 0.4 hours N = XR = 8 customers
29
If you did not get it The 120 customers sent a total of 120×24 customer×minutes or 48 customer×hours in the parlor 48 customer×hours/6 hours = 8 customers Same as having 8 customers spending six hours each inside the parlor
30
Network of servers (I) Arrivals Departures Open network
31
Network of servers (II) Arrivals Departures Closed network
32
Operational Quantities Keep same quantities as before but add indices 0 for whole system k for individual servers Two changes We never care about the utilization of the whole system We add number of visits V k of each server
33
Operational quantities Over the observation period, we measure C = the number of job completions C k = the number of tasks completed by device k We define X 0 = C/T = the system throughput X k = C k /T = the output rate at server k V k = C k /C = the visit count at server k
34
Important relationships C k = V k C Since each job requires V k visits, there are V k more server completions than job completions X k = V k X 0 Same property applies to throughputs
35
System response time (I) We define Nbar = average number of jobs in the system nbar i = average number of jobs at device i Nbar = Σ i nbar i
36
System response time (II) Applying Little’s law, we have R = Nbar/X 0 and nbar i = R i X i = R i V i X 0 Hence R = Σ i V i R i
37
Note This result is trivial The total time spent by a job in the system is the sum of the times spent at each server This includes the time spent waiting in the server queues
38
Problem 1 A job requires 100 ms of CPU time 9 disk accesses Each disk access takes 7 ms We want V CPU and S CPU
39
Answer We now that jobs get CPU first and last V CPU = 10 Then S CPU = 100/10 =10s
40
Bottleneck analysis (I) A system has one CPU and one disk drive It processes transactions such that V CPU = 12 and S CPU = 5ms V Disk = 11 and S DISK = 8ms What is the maximum system throughput?
41
Bottleneck analysis (II) We compute first the maximum device throughputs Maximum X CPU = 1/0.005 = 200 requests/s Maximum X disk = 1/0.008 = 125 requests/s Since X i = V i X 0 Maximum throughput compatible with CPU workload is 200/12 = 16.7 transactions/s Maximum throughput compatible with disk workload is 125/11 = 11.4 transactions/s
42
Bottleneck analysis (III) The disk is this the bottleneck It has highest V i S i product Identifying feature of any bottleneck device Increasing the system throughput might require Sharing disk requests with a second disk Increasing the efficiency of the system I/O buffer
43
Problem 2 In the previous example, which device was the bottleneck? What would be the throughput of the system if the bottleneck utilization was 80%?
44
Answer We compare V CPU S CPU V disk S disk
45
Answer We compare V CPU S CPU = 100ms V disk S disk = 9×7 = 63 ms The CPU is the bottleneck
46
Answer If the bottleneck was operating at 100% utilization, It could process one job each V CPU S CPU time units Or 1/(V CPU S CPU ) job per time unit At U CPU utilization, It will process U CPU /(V CPU S CPU ) job per time unit
47
Answer X 0 = U CPU /(V CPU S CPU ) = 0.80/0.10 seconds 8 jobs/second
48
Systems with terminals M Terminals Whole system
49
Interactive response time formula We have M terminals Think time Z between the completion of a job and the submission of the next job Applying Little’s law to the whole system M = (R + Z ) X 0 then R = M/X 0 – Z Very Important
50
Problem 3 We have M = 50 users Z = 20 s X 0 = 2 transactions/s What is the system response time?
51
Answer We apply R = M/X 0 – Z
52
Answer We apply R = M/X 0 – Z and obtain R = 50/2 – 20 = 5 seconds
53
Problem 4 A system Processes 5 transactions/seconds Has 60 users Achieves a response time of 4 seconds What is the think time?
54
Answer We apply R = M/X 0 – Z, Z = M/X 0 – R
55
Answer We apply R = M/X 0 – Z, Z = M/X 0 – R = 60/5 – 4 = 8 seconds
56
Problem 5 We have M = 50 users Z = 20 s R = 4 s What is the system throughput?
57
Answer From R = M/X 0 – Z, we have X 0 = (R + Z)/M Hence X 0 = (20 + 4)/50 = 0.48 tasks/s
58
Problem 6 A system Can process up to 4 transactions/second Has 60 users User think time is 12 seconds Can the system achieve a response time of 2 seconds?
59
Answer Applying R = M/X 0 – Z, we compute a lower bound for the response time R min = M/X 0,max – Z
60
Answer Applying R = M/X 0 – Z, we compute a lower bound for the response time R min = M/X 0,max – Z = 60/4 – 12 = 3 seconds Answer is no
61
Problem 7 Compute the response time of a system knowing the following parameters M = 50 users Z = 15 s V CPU S CPU = 200ms U CPU = 50%
62
Answer Since X k = U k /S k and X k = V k X 0, X 0 = U k /(V k S k ) The response time is then given by R = M/X 0 – Z
63
Answer Let us compute first the throughput X 0 Applying X 0 = U k /(V k S k ) X 0 = 0.50/0.200 = 2.5 interactions/s The response time is then R = M/X 0 – Z = 50/2.5 – 15 = 5 s
64
Simulation Case Studies
65
A simple reminder If interarrival times are Independent identically distributed (i. i. d.) According to an exponential law then the probability of having exactly n arrivals during a fixed interval is distributed according to a Poisson law
66
Explanation (II) Assume that The probability of one arrival during a small interval t is t The probability of two arrivals during the same small time interval is negligible tt tt tt tt tt tt
67
Explanation (I) The probability of having exactly k arrivals during n slots is What would happen if the number of time intervals goes to infinity while their total duration T = n t remains constant
68
Explanation (III) We rewrite the previous expression as and compute separately the limits of its four factors
69
Explanation (IV)
70
Explanation (V) We obtain the Poisson distribution The probability that there are no arrivals in the same time interval T (or in any time interval T ) is
71
Explanation (VI) This last expression is the probability that the time interval between two consecutive arrivals is greater than T The probability that the time interval between two consecutive arrivals is equal or lesser than T is which is the cdf of the exponential distribution
72
A final observation Use the Poisson distribution to generate number of arrivals during a time interval Use the exponential distribution to generate interarrival times
73
Linear Regression
74
Most important point Compute a regression line Compute regression coefficient
75
Example
76
Linear Regression We have one independent variable One dependent variable We must find Y = + X minimizing the sum of squares of errors i (y i - - x i ) 2
77
Formulas
78
Calculations (I)
79
Calculations (II)
80
Outcome
81
More notations
82
More notations (II) Solution can be rewritten
83
Coefficient of correlation r = 1 would indicate a perfect fit r = 0 would indicate no linear dependency
84
Calculations
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.