Date post: | 04-Jun-2015 |
Category: |
Technology |
Upload: | claude-falguiere |
View: | 2,523 times |
Download: | 0 times |
YAJUG - Octobre 2012Claude Falguière @cfalguiere goo.gl/4mmJQ
DES TESTS DE PERFORMANCES ...MAIS POURQUOI ?
DES TESTS DE PERFORMANCES ...MAIS POURQUOI ?
mercredi 7 novembre 12
CLAUDE FALGUIERE
@cfalguiere
Technical architect
http://cfalguiere.wordpress.com
Duchess France JUG Leader
+Claude Falguiere
mercredi 7 novembre 12
performance
how to persuade your boss
design relevant tests
be aware of bias
prepare for the go live
mercredi 7 novembre 12
marathon
Performance ?
sprintou
mercredi 7 novembre 12
speedcapacity
volumeor
or
mercredi 7 novembre 12
and availability
mercredi 7 novembre 12
and efficiency
mercredi 7 novembre 12
2 s 6 s 8 sattente constaté
distance between happy and unhappy
It can be measured ...
mercredi 7 novembre 12
assumed complexity
rendering order
performance stability
... and it depends on users
mercredi 7 novembre 12
We use the most performing frameworks !
Let’s upgrade the CPU
Let’s go to the Cloud
1
2
3
mercredi 7 novembre 12
Performance enhancement
costsrevenues
Faster
Amazon :+100 ms -1% sales
Google : +500 ms –20% pages seen
mercredi 7 novembre 12
Less calls to the call centerLess customers in the office
savings
Performanceenhancement
costs
mercredi 7 novembre 12
savings
Less resources
Performance enhancement
costs
mercredi 7 novembre 12
Performance enhancement
temps
costs
revenues
More turnover
Less costs
mercredi 7 novembre 12
An endless struggle
mercredi 7 novembre 12
Individual Group Crowd
Behavior under stress
mercredi 7 novembre 12
Crowd
«unit» Test
breakdown
Individual Group
Slowness
Iterations
Volume
Wait and timeouts
mercredi 7 novembre 12
YSlow Page Speed
HTTP Proxy
mercredi 7 novembre 12
mercredi 7 novembre 12
Crowd
mimic a few concurrent users
Individual Grouprace conditioncontentionlive/dead locks
mercredi 7 novembre 12
Foule
Load test Peak stress test
Individu GroupeCapacity bounds
Overloading Leaks
Aging test Failure simulation
iso-prod
mercredi 7 novembre 12
Sizing according to the capacity
mercredi 7 novembre 12
REQ-1234 : High performance is required
mercredi 7 novembre 12
Who ?
What ?
Where ?
When ?
How ?
Why ?
mercredi 7 novembre 12
Essential
Frequent
High-Risk
FrequentFrequent
FrequentWhat are they doing ?
mercredi 7 novembre 12
mercredi 7 novembre 12
mercredi 7 novembre 12
REQ-5678 : Landpage should show up in less than 600 ms
mercredi 7 novembre 12
Time Serie
and distribution
mercredi 7 novembre 12
Moyenne 1,2s 90% des relevés sont < 3,5s
Unaware of
StrikingRare
Quantile 90% = 3,5s
mercredi 7 novembre 12
Some bad response time
Unstable response time
Bimodal !? ...
mercredi 7 novembre 12
Goals and drivers
Indicators
Scenarios
Volumes
Datasets
Types of test and duration
Target platform
mercredi 7 novembre 12
Garbage in - Garbage outrepresentativeness
mercredi 7 novembre 12
bias
mercredi 7 novembre 12
Find biases to improve the result
Find biases to degrade the result
mercredi 7 novembre 12
volumesscenario, data, search phrases
cachestest duration and rate
ignore errorsignore uncommon use cases
SimplifyArrange
mercredi 7 novembre 12
Which show better results ?
A B
mercredi 7 novembre 12
Which shows more errors ?
A B
mercredi 7 novembre 12
Iso-prod not required
algorithmscontentionsrobustnessscalability
agingfailure simulation
Anticipate
Iso-prod
load test (capacity)response time
reliability on load
mercredi 7 novembre 12
Design for failure
mercredi 7 novembre 12
Test to learn
mercredi 7 novembre 12
Prepare monitoring
mercredi 7 novembre 12
Claude Falguiè[email protected]/4mmJQ
Thank you
"Reality is merely an illusion, albeit a very persistent one. "- Albert Einstein
mercredi 7 novembre 12