Slide 1

Slide 1 text

GRAPHS

Slide 2

Slide 2 text

WILL FARRINGTON S @WFARR W OPS

Slide 3

Slide 3 text

LESSONS LEARNED

Slide 4

Slide 4 text

LESSONS LEARNED WITH SCIENCE!

Slide 5

Slide 5 text

GRAPHS YOU NEED TO HAVE

Slide 6

Slide 6 text

$/MIN

Slide 7

Slide 7 text

HOW MUCH MONEY DOES DOWNTIME REALLY COST?

Slide 8

Slide 8 text

HIGH AVAILABILITY IS GREAT... IF YOU NEED IT.

Slide 9

Slide 9 text

USE DATA. INFORM YOUR DECISIONS.

Slide 10

Slide 10 text

FORECASTED GROWTH

Slide 11

Slide 11 text

YOUR THINGS, UTILIZATION, AND ANYTHING THAT CHANGES

Slide 12

Slide 12 text

IT’S NOT A PROBLEM UNTIL IT’S A PROBLEM

Slide 13

Slide 13 text

BUT

Slide 14

Slide 14 text

THEN IT’S A PROBLEM

Slide 15

Slide 15 text

SO PLAN TO HAVE PROBLEMS

Slide 16

Slide 16 text

PLAN TO DEAL WITH PROBLEMS

Slide 17

Slide 17 text

WHEN THEY’RE ACTUALLY PROBLEMS

Slide 18

Slide 18 text

HOW YOU SHOULD MEASURE

Slide 19

Slide 19 text

MEAN

Slide 20

Slide 20 text

STDEV

Slide 21

Slide 21 text

WHAT IS YOUR DISTRIBUTION OF DATA RELATIVE TO THE MEAN?

Slide 22

Slide 22 text

VARIANCE

Slide 23

Slide 23 text

HOW DIFFERENT ARE YOUR DATA POINTS FROM THE MEAN?

Slide 24

Slide 24 text

LOTS OF VARIANCE IMPLIES UNPREDICTABILITY

Slide 25

Slide 25 text

IT'S OK IF A QUERY IS SLOW AS LONG AS IT IS ALWAYS SLOW

Slide 26

Slide 26 text

IT'S OK IF A QUERY IS SLOW AS LONG AS IT IS ALWAYS SLOW FACEBOOK ENGINEERING

Slide 27

Slide 27 text

PERCENTILE

Slide 28

Slide 28 text

95 98 99 99.9

Slide 29

Slide 29 text

WHAT ARE YOUR GREATEST OUTLIERS?

Slide 30

Slide 30 text

QUERIES/SECOND MEANS NOTHING IF YOUR EDGE CASES SUCK

Slide 31

Slide 31 text

QUERIES/SECOND MEANS NOTHING IF YOUR EDGE CASES SUCK FACEBOOK ENGINEERING

Slide 32

Slide 32 text

WHY YOU SHOULD MEASURE

Slide 33

Slide 33 text

DATA SHOULD INFORM YOUR DECISIONS

Slide 34

Slide 34 text

FEELING DRIVEN DEVELOPMENT

Slide 35

Slide 35 text

THIS THING “FEELS” FASTER

Slide 36

Slide 36 text

THIS THING “FEELS” FASTER NO

Slide 37

Slide 37 text

DATA DRIVEN DEVELOPMENT

Slide 38

Slide 38 text

MONITOR MEASURE MANAGE

Slide 39

Slide 39 text

DATA DRIVEN DEVELOPMENT YES

Slide 40

Slide 40 text

IT’S LIKE THE SCIENTIFIC METHOD ON REPEAT

Slide 41

Slide 41 text

No content

Slide 42

Slide 42 text

No content

Slide 43

Slide 43 text

lulz lullingly provided by @techpickles

Slide 44

Slide 44 text

TOOLS

Slide 45

Slide 45 text

NEWRELIC

Slide 46

Slide 46 text

SCOUT

Slide 47

Slide 47 text

METRIKS https://github.com/eric/metriks

Slide 48

Slide 48 text

LIBRATO METRICS

Slide 49

Slide 49 text

(STATSD) + GRAPHITE

Slide 50

Slide 50 text

WILL FARRINGTON S @WFARR W OPS speakerdeck.com/u/wfarr

Slide 51

Slide 51 text

THANKS