Great Mobile Performance Innovations for 2013
By Product Management | January 7, 2013
Web Performance is something many of us deal with daily as we try to make our web sites work faster for our users. There is no need to review why "faster is better", this is an axiom that we all know to be true.
If you follow the topic of 'web performance' then undoubtedly you know about Steve Souders and his recent work in mobile performance which is an extension of his ongoing work on traditional web performance.
You can see some of Steve's work and research at his site: http://stevesouders.com/mobileperf/
I'd like to focus on one of the tools listed in his toolbox section:
With 'pcapperf' you connect your mobile device over Wifi to another computer on your local network, then you capture the raw packets flowing over that Wifi connection to your handset.
This is great in that you can produce a classic waterfall diagram of a web session. Go to http://http://pcapperf.appspot.com/ to see an example output.
Now let's take it to the next level. Instead of connecting your mobile device over Wifi to another local computer, what if you wanted to produce a waterfall chart of your service talking over the mobile operators real RF network? I'm not talking about Wifi at all, but instead the 3G, 4G, or LTE connection used by your device to talk to the carrier cell towers.
Pcapperf doesn't use the default connection to the carrier network and instead performs the packet capture over local Wifi. This is still "great" in all respects, but you could argue that this doesn't really represent the true mobile experience of the masses since most of the time we aren't connected over Wifi while using our mobile devices. It is nice to see a waterfall diagram of handset traffic over Wifi, but what about low level mobile traffic over AT&T, T-Mobile, Verizon, Sprint, and the other carrier networks?
Keynote has taken the pcapperf process one step further. It is now possible to capture low level network traffic over the operators network connection (not Wifi), and still produce a waterfall diagram using the tools on the pcapperf site. Or you can just look at the data in WireShark or similar packet analysis tool.
The number one thing you might do with this is compare mobile Wifi performance versus the performance of content over the operators network. If some of your content takes 3 seconds to load as reported by pcapperf over Wifi, but it takes 14 seconds to load over AT&T, Sprint, or Verizon, would that be important to know?
Here is a waterfall diagram created by capturing low level packet information right from a real handset which is talking over the AT&T LTE network:
You can tell this snippet of a full waterfall diagram is from the eBay mobile site. The times represent what it takes to load the eBay mobile home page over the AT&T LTE network, and not over Wifi. The data was also collected directly from a real mobile handset, in this case an Android device. (for brevity, the entire waterfall diagram is not shown)
This innovation is a great step forward in mobile performance analysis. It is one step closer to measuring content in the context of the "true mobile experience". Pcapperf taken to the next level!
This also works for mobile "apps" in addition to mobile web, and you can capture across multiple "steps" and aren't limited to a single URL. If you have a mobile app or mobile web site, you can capture "Login", "Search", "Add-to-Cart", "Check-out", etc. Then analyze all of the low level network traffic for each step in your service.