"Knowing" isn't the same as measuring

[Tim Holloway]

We often advise people in this forum not to waste time in prematurely optimizing the parts of their apps that they "know" need optimizing. Here's a real-world example of why:

http://www.wired.com/epicenter/2011/03/android-iphone-web/

[Ulf Dittmer]

The methodology doesn't convince me. Using a mix of handsets -and not even saying which ones in the case of iOS- just muddles the numbers.

The Nexus S and the Galaxy S have 1 GHz ARM Cortex A8 CPUs.

iPhone 4.2 and 4.3 are not actual devices but OS versions that run on a) the iPhone 3 on a 412 MHz ARM 1176, b) the iPhone 3GS on a 600 MHz ARM Cortex-A8, and c) the iPhone 4 on a 800 MHz ARM Cortex-A8 (http://en.wikipedia.org/wiki/IPhone#Model_Comparison).

So at least 25% difference (800MHz vs. 1GHz) can be chalked up to the CPU, if not 66% (600MHz vs. 1GHz), assuming that the iPhone 3 wasn't actually used. The latter case would negate the "findings" of this comparison. So while a comparison of a single device running iOS vs. a single device running Android might make sense, this test says nothing about iOS performance vs. Android performance.

[Tim Holloway]

I think rather than the merits of hardware or OS, the paragraph of note in this article is the final one. And, there's at least some reason to believe that the real bottleneck wasn't the devices, but the efficiency of the pipes feeding the devices.

My own experience, BTW, has been that the overhead of downloading an extensive AJAX support package can easily dwarf everything else. In my case, simply bundling the AJAX libraries into a single unit and enabling client-side caching meant a full order of magnitude in speed improvement.