methodology easy but important 1. toc 1.what is performance evaluation about ? 2.metrics, load and...

48
Methodology easy but important 1

Post on 19-Dec-2015

212 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Methodology

easy but important

1

Page 2: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

ToC

1. What is performance evaluation about ?2. Metrics, Load and Goals

3. Hidden Factors4. The Scientific Method

5. Patterns

2

Page 3: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

What is Performance Evaluation ?

Page 4: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

LoadYou need to define the load under which your system operatesMake the difference between

Intensity of the load (e.g. nb jobs per second)Nature of the load

Statistical details that may matter: e.g. job sizes are heavy tailed or not

Benchmarks are artificial load generators; we will play with one of them

Page 5: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

MetricDefine a metric; examples

Response timePower consumptionThroughput

Define operational conditions under which metric is measured (« Viewpoint », see Chapter 11)

Page 6: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Compare Windows vs Linux

6

Page 7: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

7

Syscall Benchmark

Page 8: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

8

Memory Access Time

Page 9: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

9

Ghostscript

Page 10: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Metrics are often Multidimensional

10

Page 11: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

A and D are non dominated

11

Page 12: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Know your goals

A1 and A3 are comparisons, A2 is an absolute statement

E2 is an engineering rule

Page 13: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

3. Hidden FactorsFactor: an element that may impact the performance

(desired factors): intensity of load, number of servers(nuisance factors): time of the day, presence of denial of service attack

13

Page 14: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

TCP Throughput Increases with Mobility

Page 15: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

TCP Throughput Decreases with Mobility

Page 16: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Why were we fooled ?Hidden factor had a more important role than the factor we were interested in

We interpreted correlation as causality

Need to be aware of all factors and incorporate in the analysis

Or randomize experiment to reduce impact of hidden factors

Page 17: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Simpson’s ParadoxA well known phenomenon -- Special case of Hidden Factor paradox when metric is success rate and factors are discrete

Page 18: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Berkeley Sex Case 1973 (source: wikipedia)

18

Page 19: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Take Home MessagePitfall number 1 is the presence of hidden factorAny study is susceptible to itEasy for opponents to find

Page 20: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

4. Be ScientificJoe measures performance of his Wireless Shop:

what would you conclude ?

Page 21: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Scientific MethodJoe buys 2 more Access Points

improvement ?

21

Before After

Page 22: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Scientific MethodA conclusion can only proven to be wrongDo not draw conclusions unless the experiment was designed to test the statement

Measurement 1 suggested that the wireless network was congested, but the experiment was not designed to test this statement

Joe should: design an experiment to validate:H1: “the wireless network is the bottleneck”

for example: measure the number of collisions / packet lossresult: collision · 1%; conclusion: H1 is not validhypothesis H2: the server is saturatedexperiment: measure memory utilization : result ¼ 100%

22

Page 23: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Performance After Doubling Server Memory

23

Page 24: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Example from Nitin Vaidya, Mobicom 2000 Tutorial, slides 298-299

24

Page 25: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

25

Page 26: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

26

Page 27: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

27

Page 28: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Use of Scientific Method

Recognize a fact

Pose a hypothesis

Verify the hypothesis on simulations / measurements designed to test it

TCP throughput may increase with mobility(1) Duration of link failure period is impacted by speed(2) It has a negative impact on TCP throughputDo more simulationsmeasure distrib of link failure period

Verify (1) and (2)How ?

28

Page 29: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Is ATM-ABR better than ATM-UBR ?

29

Page 30: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Take Home MessageYou should not conclude from an experiment without trying to invalidate the conclusion

(Popper, 1934): you should alternate between the roles ofProponentAdversary

Page 31: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

5. Patterns

These are common traits found in different situationsKnowing some of them may save a lot of time

Page 32: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns
Page 33: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Bottlenecks may be your enemyBottlenecks are like non invited people at a party – they may impose their agenda

Previous example: what we are measuring is the bottleneck, not the intended factor

33

Page 34: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Bottlenecks are Your Friends

Page 35: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Bottlenecks are Your FriendsSimplify your life, analyze bottlenecks !In many cases, you may ignore the rest

35

Page 36: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Behind a Bottleneck May Hide Another Bottleneck

Page 37: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

37

i’

i”

i

i”

Page 38: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

i’

i”

i

i”

38

Page 39: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

39

Page 40: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Congestion CollapseDefinition: Offered load increases, work done decreasesFrequent in complex systemsMay be due to

cost per job increases with loadImpatienceRejection of jobs before completion

Designer must do something to avoid congestion collapseEg. Admission control in Apache serversEg. TCP congestion control

Analyst must look for congestion collapse

40

Page 41: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Sources use TCP (= fair scheduling). Increase capacity of link 5 to 100 kb/s; what happens ?

41

Page 42: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Competition Side EffectSystem balances resources according to some schedulingApparent paradox: put more resources, some get less

42

Page 43: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

No TCP, users send as much canIncrease capacity of link 2 from 10 to 1000 kb/s

43

10 kb/s

Page 44: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Competition Side EffectApparent paradox: put more resources, all get less

44

Page 45: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Museum Audio Guide

45

Low speed USB connections at docking stationHigh speed

Page 46: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Latent Congestion CollapseSystem is susceptible to congestion collapseLow speed access prevents congestion collapseAdding resources reveals congestion collapse

46

Page 47: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Take Home MessageWatch for patterns, they are very frequent

BottlenecksCongestion collapseCompetition side effectsLatent Congestion collapse

47

Page 48: Methodology easy but important 1. ToC 1.What is performance evaluation about ? 2.Metrics, Load and Goals 3.Hidden Factors 4.The Scientific Method 5.Patterns

Now it’s your turn…

48