Understanding Mobile Web Performance
- 1. Understanding Mobile Web Browser Performance Rajiv Vijayakumar Qualcomm Incorporated QUALCOMM CONFIDENTIAL AND PROPRIETARY 1
- 2. Topics § Mobile browser architecture § Mobile network latencies § Android HTTP: Parallel downloads and caching § Power consumption 2
- 4. What is a “WebKit” Browser? 4
- 6. Browser Architecture Layout Engine 5
- 7. Browser Architecture WebKit 5
- 8. Browser Architecture JavaScript WebKit Engine 5
- 9. Browser Architecture JavaScriptCore WebKit or V8 5
- 10. Browser Architecture JavaScriptCore WebKit or V8 Rendering Networking 5
- 11. Browser Architecture Browser UI JavaScriptCore WebKit or V8 Rendering Networking 5
- 12. Android 6
- 13. Android Browser ≠ Chrome 7
- 16. <img src="http://www.qualcomm.com/imgA.jpg"> DNS Time 8
- 17. <img src="http://www.qualcomm.com/imgA.jpg"> DNS Time 96.16.97.9 www.qualcomm.com 8
- 18. <img src="http://www.qualcomm.com/imgA.jpg"> DNS TCP Time 8
- 19. <img src="http://www.qualcomm.com/imgA.jpg"> DNS TCP Time SYN SYN/ACK 8
- 20. <img src="http://www.qualcomm.com/imgA.jpg"> DNS TCP HTTP Time 8
- 21. <img src="http://www.qualcomm.com/imgA.jpg"> DNS TCP HTTP Time HTTP/1.1 200 OK GET /imgA.jpg 8
- 22. <img src="http://www.qualcomm.com/imgA.jpg"> DNS TCP HTTP Time HTTP/1.1 200 OK GET /imgA.jpg 8
- 23. Latencies in 3G/4G Networks 9
- 24. Latencies in 3G/4G Networks § Initial connection setup: > 1s (on older 3G) 9
- 25. Latencies in 3G/4G Networks § Initial connection setup: > 1s (on older 3G) § Round trip time (RTT): 80ms or higher (typically) § 100 - 200ms common 9
- 26. RTTs at speed of light § San Diego – New York : ~26ms § San Diego – Johannesburg : ~110ms 10
- 27. Every round trip is at least 80-100ms DNS TCP HTTP 11
- 28. Every round trip is at least 80-100ms DNS TCP HTTP Including HTTP retrievals from a CDN 11
- 29. DNS Every new lookup takes 1 RTT + Lookup Time Histogram of DNS Lookup Times 150 100 50 0 0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 Seconds 12 Source: Data collected over 30 sites (280 domains) on Qualcomm corporate network
- 30. Domain Sharding § Use of multiple domains (cdn1.domain.com, cdn2.domain.com) § Requests may not happen in parallel... 13
- 31. Parallel Downloads 8 images across 2 domains cdn1.domain.com/img1A.png cdn2.domain.com/img2A.png cdn1.domain.com/img1B.png cdn2.domain.com/img2B.png cdn1.domain.com/img1C.png cdn2.domain.com/img2C.png cdn1.domain.com/img1D.png cdn2.domain.com/img2D.png 14
- 32. Browsers typically use 4-6 sockets per host img1A img1B img1C cdn1 img1D img2A img2B img2C cdn2 img2D 15 t=0
- 33. Android img1A img1D img1B img1C cdn1 img2A img2B img2C img2D cdn2 16
- 34. Android img1A img1D img1B img1C cdn1 img2A img2B img2C img2D cdn2 16
- 35. Android img1A img1B img1C cdn1 img2A cdn2 17
- 36. Android img1A § “Thread per socket” model img1B cdn1 img1C § 4 HTTP threads img2A cdn2 17
- 37. QUALCOMM CONFIDENTIAL AND PROPRIETARY 18 Source: DDMS (from Android SDK) connected to Android 2.3.3 emulator
- 38. Android img1A § “Thread per socket” model img1B cdn1 img1C § 4 HTTP threads img2A cdn2 19
- 39. Android img1A § “Thread per socket” model img1B cdn1 img1C § 4 HTTP threads § Only 4 sockets can have img2A outstanding HTTP requests cdn2 (“active”) 19
- 40. Android img1A img1D img1B img1C cdn1 img2A img2B img2C img2D cdn2 20
- 41. Android img2A img2B img2C img2D cdn2 20
- 42. Android img2A img2B img2C img2D cdn2 Pipelining 20
- 43. Android img2A img2B img2C img2D cdn2 Pipelining § Send multiple requests without waiting for each response § Responses must arrive in order 20
- 44. Android img2A img2B img2C img2D cdn2 Pipelining § Send multiple requests without waiting for each response § Responses must arrive in order § Android pipelines up to 3 requests on each socket 20
- 45. Pipelining § Not supported, or not enabled by default on most major browsers § Historically caused problems (e.g. due to badly behaving proxies) § Millions of Android devices actively using pipelining § Time to revisit... ? 21
- 46. HTTP Caching 22
- 47. HTTP Caching § WebKit maintains a memory cache § Default size = 8 MB § Android has a persistent (file system) HTTP cache § Default size = 6 MB 22
- 49. Cache Eviction § What gets evicted to make room? 23
- 50. Cache Eviction § What gets evicted to make room? § Android policy: Prioritize base on expiration date § Far future expiration dates have higher priority § Evict lowest priority items first Expires:Wed, 09 Jun 2021 21:12:36 GMT 23
- 51. Example of Cache Contents Visited several sites, read blog posts and linked articles etc. 24
- 52. Example of Cache Contents Visited several sites, read blog posts and linked articles etc. sqlite3 webviewCache.db 'select expires from cache order by expires desc;' 24
- 53. Example of Cache Contents Visited several sites, read blog posts and linked articles etc. sqlite3 webviewCache.db 'select expires from cache order by expires desc;' Expiration Year 2036 2035 ... 2011 24
- 54. Example of Cache Contents Visited several sites, read blog posts and linked articles etc. sqlite3 webviewCache.db 'select expires from cache order by expires desc;' 200 Expiration Year 150 2036 Count 2035 100 ... 50 2011 0 2010 2020 2030 2040 24 Expiration Year
- 55. § Cache eviction policy affects performance § Consider real-world cache occupancy when measuring cached page load performance 25
- 56. Caveats § Devices can (and do!) differ from stock Android § Differences in: § Number of HTTP threads § Cache size and eviction algorithm § etc. 26
- 57. Power What is the power drawn from the battery while loading a web page? 27
- 58. Battery Current (mA) 28 Source: Power measurements made on MDP MSM8655™ running Android 2.3 over commercial 3G network
- 59. 11s download (400mA) Baseline (200 mA) 10s idle (300mA) QUALCOMM CONFIDENTIAL AND PROPRIETARY 29 Source: Power measurements made on MDP MSM8655™ running Android 2.3 over commercial 3G network
- 60. Power § Any network traffic brings up the radio for several seconds § Network dormancy timer: typically 10-15 seconds § Use long periods for any periodic Ajax calls etc. 30 Source: Power measurements made on MDP MSM8655™ running Android 2.3 over commercial 3G network
- 61. TCP FINs and Power 31 Source: Power measurements made on MDP MSM8655™ running Android 2.3 over commercial 3G network
- 62. TCP FINs and Power § Android closes TCP sockets after ~6s of inactivity § Sockets get closed at different times 31 Source: Power measurements made on MDP MSM8655™ running Android 2.3 over commercial 3G network
- 63. TCP FINs and Power § Android closes TCP sockets after ~6s of inactivity § Sockets get closed at different times § Every socket close generates a TCP FIN § Extends the dormancy timer § Keeps the modem on for a 1-bit communication 31 Source: Power measurements made on MDP MSM8655™ running Android 2.3 over commercial 3G network
- 64. TCP FINs and Power § Android closes TCP sockets after ~6s of inactivity § Sockets get closed at different times § Every socket close generates a TCP FIN § Extends the dormancy timer § Keeps the modem on for a 1-bit communication § Optimization: close all sockets after page load completes 31 Source: Power measurements made on MDP MSM8655™ running Android 2.3 over commercial 3G network
- 65. JavaScript Performance § SunSpider benchmark numbers have improved dramatically in recent years § Android (Jun 2011) < 2.5s § Desktop Chrome (Sep 2008) ~2.8 s 32 Source: http://ejohn.org/blog/javascript-performance-rundown/
- 67. Q&A 34
- 68. Disclaimer QUALCOMM Incorporated 5775 Morehouse Drive San Diego, CA. 92121-1714 U.S.A. Copyright © 2011 Qualcomm Incorporated. All rights reserved. Not to be used, copied, reproduced in whole or in part, nor its contents revealed in any manner to others without the express written permission of Qualcomm. Qualcomm and MSM8655 are trademarks of Qualcomm Incorporated, registered in the United States. Other product and brand names may be trademarks or registered trademarks of their respective owners. Nothing in these materials is an offer to sell any of the components or devices referenced herein. Certain components for use in the U.S. are available only through licensed suppliers. Some components are not available for use in the U.S. 35