PageSpeed Minification Pecentages Miscalculated

It looks like PageSpeed is miscalculated the percent reduction for the JS Minify task.

I set up a small web page that loaded jquery 1.10.2 (non-minified) as a test.

Here’s the result from WPT:
Minifying could save 38.8KiB ([color=#FF0000]15% reduction[/color]) after compression.

Here’s the result using the current version of the PageSpeed Chrome extension:
Minifying could save 38.8KiB ([color=#FF0000]50% reduction[/color]) after compression.

As a control I hit both the non minified and minified versions hosted on google (gzip compressed) and looked at the sizes via Fiddler: (Gzip, not-minified): 79.2KiB (Gzip, minified): 32 KiB
Reduction: 47 KiB (60%)

So the Chrome extension version seems much more accurate, I’m guessing the remaining difference is due to various differences in the gzip and/or minfication algorithms used.

Is this just a matter of updating the PageSpeed version in WPT or am I missing something?


FWIW, I’m actually in the process of removing the pagespeed lib and using the online service anyway. Both the extension and the lib have been falling further behind (the lib hasn’t been updated in several years).

Thanks for the info, would there be a solution for private instances?

Not a good one (at least for private pages). For public pages you’ll be able to use a developer key and also integrate with Page Speed but nothing that runs behind a firewall.

Are there specific checks you are interested in? WebPagetest does it’s own checks for some of the basic things. It used to also do a minify check but after gzip I wasn’t seeing it matter that much so I had disabled it.

That’s unfortunate, we’re trying to implement the private instance specifically for our intranet pages.

Probably the biggest checks we’re interested are the minify tasks as it’s not as easy to replicate as some of the other checks with just the results from WPT. I think there’s still a significant savings of minifying + gzip over just gzip alone as shown from the jquery example above (60% savings).

There’s others that I think are useful, the render blocking and above-the-fold checks. It seems like there’s enough data from WPT to replicate these but to be honest I don’t know enough about the algorithms.