Measuring Mobile Web Performance presentation at the London Ajax Mobile Conference 2nd July 2011. Covers the basics of web performance measurement and looks specifically at the measurement of page load speed from mobile devices.
Here is one reason why a mobile browser isn’t the same as a desktop browser….(on Android) the “web browser” is a Java App calling a C/C++ app which calls another C app…
Different bandwidthDifferent latencyDifferent network layer entirely (Backhaul probably isn’t even IP-based…)
And if you throw in CyanogenMod you can get even better performance…
And if you throw in CyanogenMod you can get even better performance…
What type of network is it connected to?
I was on the Thames Clipper on my way back from the O2 after a concert. So I ran some automated testing and this graphs my results… Average32.859Median12.071Std Deviation45.691Min9.275Max156.520
“Snakes… on a motherf**king plane!” Network latency varies enormously…. What can we see here?The connection is relatively good at sending and receiving data. The green crosses represent packets that were actually lost which only happened for 75 of 1000.There are some crazy-high round trip times. The minimum round trip time was 107ms (which would put my home cable connection to shame) and even the median is pretty awesome at 239ms but the maximum was a whopping 20226 ms - that's more than 20 seconds!
Test MarkerTime (Sec)Signal Strength (ASU)19.2753212.681339.68274156.52018512.071969.28513752.86814810.1849923.1658Average32.859Median12.071Std Deviation45.691
1.JavaScript timing e.g. WebTuna, Boomerang, Episodes2.Browser plug-in e.g. HTTPwatch3.Custom browser variant e.g. A Webkit-build (often used as a monitoring agent)4.Proxy timing e.g. Fiddler proxy5.Web Server Module e.g. APM solutions like AppDynamics, Dynatraceetc6.Network-level e.g. Atomic Labs Pion, Tealeaf, Coradiant
2.Browser plug-in e.g. HTTPwatchBut only some mobile browsers have a plug-in architecture… and exactly what timing info that would provide… I have no idea…
4.Proxy timing e.g. Fiddler proxy or Charles ProxyObserver Affect #1 = http://www.fiddler2.com/fiddler/perf/PerfMeasure.aspObserver Affect #2= http://insidehttp.blogspot.com/2005/06/using-fiddler-for-performance.html
5.Web Server Module e.g. APM solutions like AppDynamics, Dynatraceetc
1.JavaScript timing e.g. WebTuna, Boomerang, EpisodesYou can also use YUI Profiler I believe http://developer.yahoo.com/yui/profilerI have seen JavaScript warnings appear due to “excessive time in the onBeforeUnload event” – so keep in mind that JavaScript is a lot slower in mobile devices and hence the observer effect is larger (and the timing more accurate).
Network tap based solutions are great for measuring your own site, not so great for measuring 3rd party sites.PCAPPerf is limited to WiFi… which isn’t anything like the real “mobile” experience. The mobile network stack is NOT the same as the desktop stack, and the client side delay is larger also. *In theory you can use tcpdump on a rooted phone. More on that later…
The mobile network stack is NOT the same as the desktop stack, and the client side delay is larger also.
Is a tethered, rooted phone over wifi adequate to your needs?
Is knowing what’s happening “on the wire” enough, or do you want to understand what’s happening at the browser level?And the “mobile context” …So lets look at a few in detail…
Flexible solution (in theory works with every WiFi enabled phone)But WiFi isn’t mobile…And it’s a faff…
Essentially the PCAPPerf approach as SaaSBut it does support iPhone and Android.
Currently Android only but PoC code for Blackberry and iPhone exists. iPhone client in 90 to 120 days, all being well.
Rooted and Tethered, again…
Even a PC-based version of Webkit isn’t the same as webkit running on a mobile device. But useful none the less. Lots of other features and unsurprisingly there is a (expensive) Pro version “The MITE Pro annual user license is $5,000 and includes 100 hours of free remote testing. To learn more about single- and multiple-license purchases, please contact us today.”
The mobile network stack is NOT the same as the desktop stack, and the client side delay is larger also.