Performance analysisclass
Download as PPTX, PDF3 likes844 views
The document summarizes Daniel Austin's presentation on "The Art and Science of Web Performance". The presentation discusses web performance as both an art and a science, and that performance is response time. It also covers dimensions of performance like geography, bandwidth, and devices. Key aspects of HTTP connections like connection time, server duration, transport time, and render time are explained. Tools for testing performance like YSlow and commercial services are also mentioned.
![T2 – THE SERVER DURATION
• Let ( ) r
• U = ( r)[ r
W]
• X =U*
• Navg =
( r [W( r)W+1 -
(W+1)( r)W+1]
• … so 2 = Navg/X
(The response
time law)
Never mind - it’s a constant!
Confidential and Proprietary 8/2/2012](https://image.slidesharecdn.com/performanceanalysisclass-120802145604-phpapp01/85/Performance-analysisclass-11-320.jpg)
![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
Confidential and Proprietary 8/2/2012](https://image.slidesharecdn.com/performanceanalysisclass-120802145604-phpapp01/85/Performance-analysisclass-12-320.jpg)
Performance analysisclass
- 1. The Art and Science of Web Performance Daniel Austin Technical Staff Paypal, Inc. CMU Invited Lecture August 02 , 2012 daaustin@paypal.com V1.1 @daniel_b_austin
- 2. TODAY’S PROGRAM Overture: Web Performance As Art And Science First the Theory… …Then the Practice Coda: Making the Web Faster and Smarter Confidential and Proprietary 8/2/2012
- 3. STATE OF THE ART? • An Art and a Science • Very Little Prior Art • Users Suffer – • The World Wide Wait isn’t over yet! • Rapid Change in the Industry • Browsers • Devices • Standards • The Rest of the World is Catching Up • Challenges of Global/Local/Mobile/Social Performance Confidential and Proprietary 8/2/2012
- 4. PERFORMANCE IS RESPONSE TIME PERFORMANCE IS RESPONSE TIME PERFORMANCE IS RESPONSE TIME PERFORMANCE IS RESPONSE TIME (It’s not latency, it’s not bandwidth, it’s not queue residence time or queue length or any such thing.) Confidential and Proprietary 8/2/2012
- 5. Who needs ‘Performance’ measurements anyway? ? Confidential and Proprietary
- 6. A MORE RATIONAL APPROACH? Confidential and Proprietary
- 7. 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 Confidential and Proprietary 8/2/2012
- 8. THE MPPC MODEL OF WEB PERFORMANCE “Multiple Parallel Persistent Connections” Request DNS/Network Initiation by HTTP Request Resolution HTTP Request User t1 This entire cycle, steps 1-4, is repeated once for each external Page reference on the page, so for a given page the total time is: Composition S T = S Dt1 + Dt2 + Dt3 + Dt4 t2 n+1 n+1 End User Where n is the number of external page requisites. Payload Browser HTTP Response Delivery Time HTTP Response Rendering Time t3 t4 T1 T2 T3 T4 Connection Time Server Duration Transport Time Render Time Confidential and Proprietary 8/2/2012
- 9. HTTP CONNECTION FLOW Client Server Connection setup (t1) Handshake time Request transmission Client’s perceived time response time Request (t2) Estimated server processing time Response transmission Response (t3) time The more HTTP requests & network roundtrips you require, the slower your site will be: Images, CSS, JS, DNS lookups, Redirects, #of packets Confidential and Proprietary 8/2/2012
- 10. T1 – MAKING THE CONNECTION 1= DNS + TCP+ SSL Confidential and Proprietary 8/2/2012
- 11. T2 – THE SERVER DURATION • Let ( ) r • U = ( r)[ r W] • X =U* • Navg = ( r [W( r)W+1 - (W+1)( r)W+1] • … so 2 = Navg/X (The response time law) Never mind - it’s a constant! Confidential and Proprietary 8/2/2012
- 12. 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 Confidential and Proprietary 8/2/2012
- 13. T4 – WHAT THE BROWSER DOES Confidential and Proprietary 8/2/2012
- 14. BANDWIDTH EFFICIENCY Bmax = 1.22*(L)^1/2 * MSS/RTT Confidential and Proprietary 8/2/2012
- 15. WHERE ARE THE DELAYS? Confidential and Proprietary 8/2/2012
- 16. LET’S TALK TOOLS Site Performance Services Page Analysis Tools – Gomez – YSlow – Keynote – MS Virtual RoundTrip – WebPagestest.Org Analyzer, • ‘Wholesale’ Testing HTTPWatch, Many Others – Statistical data for many page views under different – F12 in your browser conditions • ‘Retail’ Testing – Operational testing – One Page or App – Best for understanding – Diagnostic global and network effects – Best for functional testing Confidential and Proprietary
- 17. 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 • You can do this for yourself on a smaller scale at WebPageTest.org Confidential and Proprietary 8/2/2012
- 18. HAPPY BIRTHDAY, YSLOW! 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 Confidential and Proprietary
- 19. TESTING YOUR COMPETITORS FOR FUN & PROFIT • Try to measure task sequences not pages • Be prepared for lots of breakage • Be respectful • Difficult and time-consuming Confidential and Proprietary 8/2/2012
- 20. TASK-BASED PERFORMANCE THINKING Welcome 1400 Inbox 2200 Inbox Bulk 3200 (B) … 61.77% P(A,B) = 0.5168 Read 10.04% 66.25% Message (D) Verify (F) 34.52% Welcome 9.85% Bulk (A) (C) 58.07% 64.53% 22.30% 34.38% 35.02% Send Compose28.04% Confirm (E) (G) Confidential and Proprietary 8/2/2012
- 21. GLOBAL PERFORMANCE TESTING (BRIEFLY) • 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 relative ‘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. Confidential and Proprietary 8/2/2012
- 22. BAKE IT IN UP FRONT Performance Engineering is a design-time activity! This is absolutely crucial! Anti-Pattern: Releasing a new or modified product without testing its performance. Seriously. Confidential and Proprietary
- 23. QUANTIFY Confidential and Proprietary
- 24. KNOW THE LANDSCAPE “What are the differences among browsers?” “What are my competitors doing?” “What about low bandwidth users? “How can I test that?” Confidential and Proprietary 8/2/2012
- 25. THE 7 HABITS OF EXCEPTIONAL PERFORMANCE 1. Make Performance a Priority 2. Test, Measure, Test Again 3. Learn about the Tools 4. Balance Performance with Features 5. Track Results Over Time 6. Set Targets 7. Ask Questions; Check It for Yourself! Confidential and Proprietary
- 26. CODA: MAKING THE WEB SMARTER AND FASTER • Faster HTTP • SPDY • HTTP Speed+Mobility • Scratch • Better Browsers • Chrome has disrupted the market • HTML5 will drive further evolution • Moving to Mobile/Device Platforms • More sensitive to network effects of all kinds • Application-driven user experience • Moving away from Hypertext? Confidential and Proprietary 8/2/2012
- 27. 42 Michael Faraday is the father of network performance, having performed the first bandwidth measurements on the first transatlantic telegraph cable in 1854. Daniel Austin Technical Staff CMU Invited Lecture Paypal, Inc. August 02 , 2012 V1.1 daaustin@paypal.com @daniel_b_austin
Editor's Notes
- #2: Note that this is a work in progressStill being worked on, things are fluid though the direction is clear
- #4: These are good problems to have.
- #7: This diagram gives you an idea of the number of different use cases and actors across a large organization who are concerned with performance and the user experience. Every one of these people needs to know something about performance.It’s really complicated in more ways than one!
- #19: Yslow turns 5 today!
- #20: Competitive testing is an art in itself
- #23: Nothing we do about performance is more important than testingEverything upfront in a disciplined way.
- #27: Heresy!