These are the slides I used for my 1-day Performance Workshop at HTML5 DevConf this year.

1. Web Performance Boot Camp
Daniel Austin
PayPal, Inc.
HTML5 DevConf
Oct. 24, 2013
V 2.2

2. Overture: Goals of the Class
• Provide a basic understanding of Web performance for
Architects, Developers, Designers, and Engineers
• Empower YOU to identify and resolve performance
problems and make your pages and applications faster!!!
• Demonstrate and explain how to use common tools and
techniques used in our industry to solve performance
problems

3. Scope of Web Performance
Anything that uses HTTP
Always From the End User‟s Point of View
Web Request/Response Only!

4. Current State of the Art
• Web performance is both an Art and a Science (but it‟s not yet
Engineering)
• Multiple tools and methodologies, large ad hoc, contend in the
marketplace (but little of it is well-thought out or based on
scientific reasoning).
• Things are getting better – W3C involvement and competitive
pressures, as well as better infrastructure and the influx of new
users in Asia is driving more attention to performance. There‟s
hope.

5. Reading List
• Performance by Design - Daniel A. Menasce (Safari)
• The Practical Performance Analyst - Neil J. Gunther
• Elements of Networking Style - M.A. Padlipsky
• High Performance Web Sites – Steve Souders

6. Tools Used in This Class
• Excel (or similar spreadsheet program)
• Online Testing Tools – webspeedtest.org
• Desktop Testing Tools – your browser, Firebug, Visual
Round Trip Analyzer (VRTA), netmon, dig, ping others
• Optional: R (and Rstudio), Mathematica, SPSS

7. Class Structure
Schedule
•
•
•
•
•
Start: 9:00 AM
Break: 10:30-1:45
Lunch: 12:30-1:15
Break: 2:30-2:45
End: 4:00 PM
Agenda
•
•
•
•
•
•
Section I – What Is Performance?
Section I – Performance Basics
Section III – The MPPC Model
Section IV – Tools & Testing
Section V – HTML5 & Peformance
Section VI – Mobile Devices

8. Section I
What is Performance?

9. What Problem Are We Trying To Solve?
• World-class response times compared to our competitors
• Reliable, predictable performance for users worldwide
• Efficient use of resources: cost scales linearly with traffic
• Delighted users!

10. Impact of Scalability on Business
• Google
– 500 ms reduces traffic to sites by 20%
• Yahoo!
– 400 ms reduces traffic y 5-9%
• Amazon
– 100 ms reduces revenue by 1%
• Compuware
– 1 sec delay reduces conversion by 7%
10

11. Who Needs Performance Anyway?
Who needs
‘Performance’
measurements
anyway?
?

12. A More Rational Approach?

13. Systemic Qualities In a Nutshell
“Anything you can say about a black box – from the outside”
• Systemic qualities are the “ilities” – physical
features of the system such as capacity,
performance, and scalability
• The SQs correspond to different groups of
stakeholders: users, developers, operators,
organizations
• SQs are the best measure of the quality of the
user‟s experience of the system, regardless of
the feature set

14. The Four Classes of Systemic Qualities
Manifest Qualities - What the users see
• Usability, Performance, Reliability, Availability, Accessibility
Operational Qualities - What the system operators see
• Throughput, Manageability, Security, Serviceability
Developmental Qualities - What developers see
• Buildability, Budgetability, Planability
Evolutionary Qualities - How the system changes over time
• Scalability, Maintainability, Extensibility, Reusability, Portability

15. The Manifest Systemic Qualities
• Usability reflects the ease with which users can
accomplish their goals
• Performance reflects how much little time users must
wait for actions to complete
• Reliability measures how often the system fails
• Availability measures uptime vs. downtime
• Accessibility measures the systems ability to serve
users regardless of location or physical condition
(including I18N and L10N)

16. Performance is a Balancing Act
Performance isn’t everything; sometimes we’re called on to
make choices about which systemic qualities have priority over
others. Security v. performance is a common tradeoff – what
would you choose?

17. What Is Performance?
PERFORMANCE IS RESPONSE TIME
PERFORMANCE IS RESPONSE TIME
PERFORMANCE IS RESPONSE TIME
PERFORMANCE IS RESPONSE TIME
PERFORMANCE IS RESPONSE TIME

18. Section II
Performance Basics

19. Scales of Measure

20. Statistics 101

21. Comparison of Mean, Median, and Mode
Comparison of mean, median and
mode of two log-normal distributions
with different skewness.

22. Outliers, or Why We Use the Median
• A: skewed to the left
• B: skewed to the right
• C: symmetrical

23. Statistical Distributions
• Discrete or continuous?
• Mean, median, sigma,
95%?
• Is it reasonable?
|mean – median| <= sigma
• Does it correlate?

24. Understanding the Margin of Error
The margin of error is a measure of how close the results are likely to be.
• Margin of error at 99% confidence
= 1.29/sqrt(n)
• Margin of error at 95% confidence
= 0.98/sqrt(n)
• Margin of error at 90% confidence
= 0.82/sqrt(n)
(where n is the number of sample
data points)

25. 5 Number Reports
A simple way of summarizing a sample
•
•
•
•
Shape of the distribution
Extreme values
Variance
Skewness
Median
1st quartile
3rd quartile
Minimum
Maximum
This is how you get a sense of the data…

26. 5 Number Reports in Excel
Excel functions:
Min = MIN(Data Range)
Q1 = QUARTILE(Data Range, 1)
Q2 = QUARTILE(Data Range, 2)
Q3 = QUARTILE(Data Range, 3)
Max = MAX(Data Range)

27. 5 Number Reports in R
Let‟s make a 5 number report in R:
http://www.r-project.org/

28. Operational Research
• Developed during WWII
for managing armies and
supply chains
• A set of rules or „laws‟ that
describe the operational
aspects of a system.
• Useful for understanding
the performance of any
system
• Utilization Law
• Forced Flow law
• Little‟s Law
• Response Time Law

29. Resources and Queues
Ri
Queue
Si
Resource
Si: Service time
Ri: Queue Residence time
i: Queue length
In general, systems consist of many combined
queues and resources

30. The Utilization Law
• The utilization (Ui) of resource i is the
fraction of time that the resource is busy.
• Xi: average throughput of queue i, i.e.
average number of requests that complete
from queue i per unit of time
• Si: average service time of a request at
queue i per visit to the resource
Ui = Xi * Si

31. Interactive Response time law
IMAGE COURTESY PROF. RAJ JAIN

32. Hands-on: Using the Response Time Law in
the Real World
Let‟s say Facebook‟s Web servers can process 10K „like‟
requests/second, and the number of concurrent users is
600K. If each user waits 5s between requests, how long will
each request take?
R
= (N/X) –Z

33. Hands-on: Using the Response Time Law in
the Real World
Let‟s say Facebook‟s Web servers can process 10K „like‟
requests/second, and the number of concurrent users is
600K. If each user waits 5s between requests, how long will
each request take?
R
= (N/X) –Z
= 6 * 10e5 / (10 requests/ms) -5 * 10e4ms
= 6 * 10e4 – 5* 10e4 ms
= 1000ms
Facebook needs more servers!

34. Antipattern: Keyhole optimization
Problem: Optimizing your project
Antipattern: Optimizing *your* project, at
the expense of everyone else!
Pattern: Your project is part of the
system – optimize for the overall system
performance, not just what you can see,
even if that means your part is lessthan-perfectly optimized.

35. Section III
The MPPC Model

36. Dimensions of Performance
•
•
•
•
•
Geography
Network location
Bandwidth
Transport Type
Browser/Device Type
• RT Varies by as much as 50%
• Page Composition
• Client-side rendering and execution effects (JS, CSS)
• Network Transport Effects
• # of Connections, CDN Use

37. Hardware and Routing

38. The OSI Stack Model

39. OSI Functionality Summary
Application
Presentation
Access network resources
Translate, compress, encrypt
Session
Establish, manage, terminate sessions
Transport
Reliable message delivery end to end
Network
Transmit packets from source to destination
Data Link
Organize bits into frames
Physical
Transmit bits

40. All People Seem To Need Data Processing
Application
All
Presentation
People
Session
Transport
Seem
To
Network
Need
Data Link
Data
Physical
Processing

41. HTTP Connection Flow
Server
Client
Connection setup
Client‟s perceived
response time
Request
Response
Handshake time
Request transmission
time
Estimated server
processing time
Response transmission
time
The more HTTP requests & network roundtrips you require, the slower your
performance will be: Images, CSS, JS, DNS lookups, Redirects, #of packets

42. The MPPC Model Of Web Performance
“Multiple Parallel Persistent Connections”
Request
Initiation by
User
S
HTTP Request
HTTP Request
t1
This entire cycle, steps 1-4, is repeated once for each external
reference on the page, so for a given page the total time is:
T = S Dt1 + Dt2 + Dt3 + Dt4
n+1
End User
DNS/Network
Resolution
Page
Composition
n+1
Where n is the number of external page requisites.
Browser
Rendering Time
HTTP Response
t4
Payload
Delivery Time
HTTP Response
t3
T1
T2
T3
T4
Connection Time
Server Duration
Transport Time
Render Time
t2

43. T1 – Making the Connection
1= DNS +
TCP+
SSL
• Typically a larger part of the E2E
than expected
• Highly variable
• SSL is slow!

44. Why DNS Matters
• Nothing happens before DNS!
• User does not see anything on their page  waiting
time
• Homework Assignment: create a host file for yourself.
Try your favorite sites without DNS!
• DNS has a great impact on user‟s perceptions in HTTP
applications

45. Interacting with DNS: dig
time

46. T2 – The Server Duration
• Let (
)
r
• U = ( r)[
r
W]
• X =U*
• Navg = (
• … so
2
r
[W( r)W+1 -(W+1)( r)W+1]
= Navg/X
(The response time law)

47. T3 – TCP Transport Time
• Single Object:
3
= Sz/R+2RTT+
idle
For persistent parallel connections:
3
= (M+1)Si/Ri+[M/kNh]*3 RTTi+
idle
… for 1 base HTML page with M objects, with Si bits, at bandwidth
Ri, k connections per host, and Nh unique hostnames

48. T4 - What the Browser Does
n
i=1
4
=
off(i)
off =
time offset
to parse the HTML, JS, CSS,
and establish the individual
connections
(to different hostnames)
4 is especially significant for mobile devices!

49. Where are the delays?

50. Bandwidth Efficiency
Bmax = 1.22*(L)^1/2 * MSS/RTT

51. Bandwidth, Latency, and All That

52. More Bandwidth? Less Latency!

53. Hands-on: Testing DNS Response
times
We‟ll use nslookup for this exercise
1. Run 10 nslookup commands for a site (e.g.
www.facebook.com)
2. Observe the response time for the DNS
lookup
3. Calculate statistics for the results
• 5 number report (summary)
• Sketch the distribution
• What can you say about the response times
for DNS?

54. Antipattern: That’s Outside My Control
It‟s never the case that there is „nothing you can do‟ about a
performance problem.
Antipattern: avoiding solving a performance issue because
you think it‟s outside your control. This path leads to
despair.
Pattern: Compensate in some other part of the E2E. Think
outside the box

56. Section IV
Tools and Testing

57. Let’s Talk Tools
Site Performance Services
– Gomez
– Keynote
– AlertSite
– ThousandEyes
• „Wholesale‟ Testing
– Statistical data for many
page views under different conditions
– Operational testing
– Best for understanding
global and network effects
Page Analysis Tools
– YSlow
– MS Virtual RoundTrip Analyzer,
HTTPWatch, Many Others
– F12 in your browser
• „Retail‟ Testing
– One Page or App
– Diagnostic
– Best for functional testing

58. Commercial Testing Services
• Gomez, AlertSite, and Keynote toolsets are similar in many ways
• Synthetic Test Setup
• Test nodes in large datacenters and/or end user‟s machines
• Statistical data about response times

59. Performance Testing Locations
Your Data Centers
QA & Test

60. HTTP Object Model
Web Page(s)
Page Objects
A Test is a sequence of one or
more URLs for which HTTP
requests will be made.
(or Components)
A Monitor is a set of
predefined Tests to be run at
specific times and places
Each Page Object has 4
associated time segments, t1, t2, t3, t4

61. Desktop Tools
• Methodology
– DOM Crawler and
Packet Sniffer
– More accurate
– Analyzes components
– Stats view
• Implements the 14 18 22
105 YSlow Rules
– All browsers except IE
– Mobile bookmarklet
– Best tool for page
analysis

62. Unix Performance Testing Tools
•
•
•
•
•
Ping
nslookup
dig
traceroute
netstat

63. Task-Based Performance Thinking
Welcome
Inbox
Bulk
…
Inbox
(B)
61.77%
Read
Message
(D)
P(A,B) = 0.5168
10.04%
66.25%
Verify
(F)
34.52%
Welcome
(A)
9.85%
Bulk
(C)
58.07%
64.53%
22.30%
Exercise: What % of users
follow the path:
A->E->F->G?
34.38%
35.02%
Compose 28.04%
(E)
Send
Confirm
(G)
1400ms
2200ms
3200ms

64. Testing Your Competitors for Fun & Profit

65. Stormcat: Global Performance Testing
•
•
Cannot compare performance data out-of-region
There are many global factors involved in performance:
• Bandwidth
• ISP
• Infrastructure
• Secular cycles (weeks, holidays, usage patterns)
•
The best approach: use the „StormCat‟ system!
• Best case (Northern California high broadband @3 AM)
• Worst Case (rural Indonesia on VSNL @ 2PM local)
• Divide the range into 5 categories equally spaced between the
best & worst: some locales will be in Cat I, some in Cat II, some in Cat
III, etc.

66. Hands-on: Analyzing Waterfall Diagrams
http://www.webpagetest.org
• Choose a location and a
browser and test:
http://www.yahoo.com

67. Cached v. Uncached

68. Antipattern: Design-time Failure
Performance is a design time activity!
Anti-Pattern: Releasing a new or
modified product without testing its
performance
“Bake it in up front!”

69. Section V
HTML5 & Performance

70. A Federated Model for HTML
Core HTML5
Canvas
2D
This is
XHTML
1.1
HTML
Markup
Web
Workers
HTML
Media
Web
Sockets
IndexDB
Web
Storage
Source: Sergey Mavrody c. 2013

71. The Co-Evolution of HTML, JS, CSS, and
XML
XML Core
XSLT
XSD
Xpath/XQuery
Document Object Model
JavaScript…
JSON…
Source: Sergey Mavrody c. 2013

72. Tower of Babel: A Problem We Have Yet to
Solve

73. The Current Browser Landscape

74. HTML5 Performance

75. W3C Resource Timing

76. W3C Navigation Timing

77. Hands-on Exercise: Testing Performance
the W3C Way
• Use the Navlet:
http://code.google.com/p/navlet/
Make a bookmark or favorite using the code

78. Antipattern: We’ll Be Done With This Soon
Performance is an ongoing activity, not fire and forget!
Antipattern: Not treating performance as a property of the
system, or only testing at release time.
Pattern: establishing a long-term performance management
plan as part of your cycle.

79. Section VI
Mobile Devices

80. The Big Picture – Mobile is Growing

81. Native Apps v. HTML5 v. Desktop
• Native Apps will run ~ 5x
faster than HTML5
• Roughly 10x slower than
desktop
• HTML5 on the mobile
device can be 50x slower
– 10x from the ARM chip
– 5x from JavaScript

82. Mobile Apps Are Slow
If you‟re designing for mobile, it‟s safe to assume you‟re
going to incur 2000ms of 3G latency.

83. Slow Compared to What?
Since 2009 mobile browsers
went from 30x to 5x slower than
desktops
– Better than Moore‟s Law
improvement (!)
– JavaScript v. Native code ~ 5x
– 4g/LTE ~ 27% faster than 3g

84. Mobile Speeds by Example

85. Users Expect More. Now.

86. Mobile JavaScript Performance
• That 5x is in the code
interpretation?
• Typed arrays
• JSON layout
• DOM Manipulation
• Garbage Collection

87. How 3G/4G Networks Work

88. 3G to 4G Migration

89. HTML5: New Features for Mobile
http://mobilehtml5.org/

90. Delay-tolerant Application Design
• Plan for offline/intermittent
connectivity
• Caching local content
– Local storage
– Don‟t be afraid to use sessions
– Use HTTP Caching headers
wisely
• Always have failure modes built-in

91. The Right Tool For the Right Job
Source: Nick Zakas

92. Best Practices for Mobile
•
•
•
•
•
•
•
Tread lightly on the JavaScript
Don‟t touch the DOM!
CDNs are less effective due to network challenges
TTFB is not a good measure of server duration
Use Web Workers for preloading
Test performance on different transport types
Test battery consumption!

93. 4 Takeaways on Mobile Performance
Mobile HTML apps are slow compared to native apps
…but it‟s not all about JavaScript
Mobile networking is a big challenge
…so design for delay-tolerance
HTML5 is designed for Mobile
…so use it (wisely)!
Use the right tool for the right job
…including the right design patterns for Mobile

94. Tools for Mobile Testing
• Speedtest/Ookla
– Variability
– Characteristics of different kinds of networks
• iCurl
– Simple HTTP Operations on your device
• HTTPWatch Basic
– Look at the Waterfall
– Gather detailed data along with iCurl

95. Hands-on Exercise: Testing Mobile
Performance

96. Antipattern: It’s The Application Stupid!
T2 (server duration ~ 35% of total E2E
– More on mobile however!
Antipattern: Failing to recognize that the distribution of the
Mobile E2E is very different from a desktop performance
profile
Pattern: Carefully analyze the MPPC numbers for your site
and identify the problems that need to be solved and in
what order.

97. Finale – Summing Up

98. The 7 Habits of Exceptional Performance
1.
2.
3.
4.
5.
6.
7.
Make Performance a Priority
Test, Measure, Test Again
Learn about the Tools
Balance Performance with Features
Track Results Over Time
Set Targets
Ask Questions; Check It for Yourself!
Thanks to Tenni Theurer

99. Yslow Rules!
• Rule 1 - Make Fewer HTTP
Requests
• Rule 2 - Use a Content Delivery
Network
• Rule 3 - Add an Expires Header
• Rule 4 - Gzip Components
• Rule 5 - Put Style sheets at the Top
• Rule 6 - Put Scripts at the Bottom
• Rule 7 - Avoid CSS Expressions
• Rule 8 - Make JavaScript and CSS
External
• Rule 9 - Reduce DNS Lookups
• Rule 10 - Minify JavaScript
• Rule 11 - Avoid Redirects
• Rule 12 - Remove Duplicate Scripts
• Rule 13 - Configure ETags
• Rule 14 - Make AJAX Cacheable
Source: Stevesouders.com

100. Every Tool Has Its Place in the Universe

101. The One Number of Truth
“42”

102. About:HTML
“…a single user-interface to many large
classes of stored information such as reports,
notes, data-bases, computer documentation
and on-line systems help”
WorldWideWeb: Proposal for a HyperText
Project
Berners-Lee & Caillau, 1990

103. Theme of the Work
Ultimately, Performance is about Respect.

104. Thank You!
Daniel Austin
PayPal, Inc.
HTML5 DevConf
Oct. 24, 2013
@daniel_b_austin
da@x.com