Xavier DÉFAGO (professor)

Tokyo Institute of Technology School of Computing

July 2020

Data Presentation

most examples adapted from:

Raj Jain. The art of computer systems performance analysis.

Guidelines for ChartsMin. Effort for Reader ‣most important aspect ‣how much effort to understand?

Example ‣ legend box ‣direct labeling ‣axes labels

eg., “Daily CPU usage” ‣ include units

eg., “CPU [seconds]” 3

System Load

Response T

ime

A

B

C

System LoadResponse T

ime

B

A

C

Better

acceptable

Guidelines for ChartsAvoid Ambiguity ‣Show coordinate axis, scale divisions, origin ‣ Identify individual curves ‣Avoid potential source of misunderstanding

4System Load

Re

sp

on

se

Tim

e

A B

BAD

Common MistakesMany Variables on One Chart ‣ saves space, but harder to read. ‣=> message lost ‣better: three different graphs

5Number of users

Response T

ime [m

s]

Throughput

Response Time

Utilization

40

30

20

10

100

75

50

25

20

15

10

5

Utilization

[%]

Throughput

[s-1]

Too many variables

BAD

Common MistakesSymbols in Place of Text ‣ symbols => readers must search text ‣ saves writers time ‣ ... or not?!

readers skip => writer's time wasted

6λ

R µ = 1

µ = 3

µ = 2

Arrival Rate

Response

time 1 job/sec

3 jobs/sec

2 jobs/secBAD Better

Common MistakesLine Chart in Place of Column Chart ‣ joining points on line chart

=> intermediate values can be interpolated

7

糖分

[mg/個

]

0

75

150

225

300

柿 桃 梨 苺En

ergy

[kJ／

100g

]

0

100

200

300

柿 桃 梨 苺VERY BAD!! Better

Pictorial GamesNonzero Origin ‣emphasize or conceal the difference

8

600

625

650

675

700

Mine

Yours

0

1000

2000

3000

4000

5000

6000

Mine

Yours 0

250

500

750

1000

Mine

Yours

BADBAD Better

Pictorial GamesDouble-Whammy Graph ‣exaggerate impact

9

Throughput

Response time

Number of users

BAD

Pictorial GamesRandom Quantities w/o Confidence Intervals ‣hides variability of the information

10

Mine

Yours

Mine

Yours

BAD Better

Pictorial GamesInappropriate Cell Size in Histograms ‣possible loss of information

11

Freq

uenc

y

0

2

4

6

8

10

12

Response time [ms]

[0,2[ [2,4[ [4,6[ [6,8[ [8,10[ [10,12[

Freq

uenc

y

02468

101214161820

Response time [ms]

[0,6[ [6,12[

BADBetter

Pictorial GamesBroken Scales in Column Charts ‣exaggerate difference ‣ same as nonzero origin

12

Res

pons

e tim

e [m

s]

0

2

4

6

8

10

12

System

A B C D E F

8

9

10

11

12

Res

ponse

tim

e [m

s]

0

1

A B C D E F

System BADBetter

Misleading Pie Chart

13https://en.wikipedia.org/wiki/Misleading_graph

BAD Better

Improper Scaling

14https://en.wikipedia.org/wiki/Misleading_graph

BAD Better