Here’s a short talk I gave earlier today.
Here’s a short talk I gave earlier today.
Here’s a talk I gave today about edu mobile web performance and related topics.
Responsive images and offline application caching do not play well together. (For an explanation, see “Gotcha #6″ of Jake Archibald’s application cache article with a moderately NSFW title.)
Pickyfill (partially) solves the problem using Scott Jehl’s picturefill as a starting point. Pickyfill stores responsive images as data URLs in LocalStorage. If your page is using the HTML5 offline Appcache, pickyfill will detect this and store picturefill images as they are loaded. It will only store the images that your device displays, so (for example) an iPhone will only cache iPhone-sized images; it will not download and store crazy large images designed for large screens.
Pickyfill makes the cached images available if the user is offline. It can also improves page load time if the user is on a slow network.
Pickyfill requires ApplicationCache, LocalStorage, and Canvas. If a browser that does not support these features visits a site that uses pickyfill, then pickyfill will do nothing and the experience will gracefully degrade to straight-up picturefill.
|Internet Explorer||10.0+||full support|
|Internet Explorer||< 10.0||degrades to regular picturefill|
|Safari (OS X, Windows)||4.0+||full support|
|Safari (iOS)||3.2+||full support|
|Android||All versions||it’s complicated**|
*In Firefox, pickyfill will cache images on load, but not on resize. This is to avoid caching truncated/corrupted images. On resize, behavior gracefully degrades to regular picturefill behavior.
**Android browser in Android 2.3 (and probably others) does not implement
toDataURL() completely/correctly. Pickyfill will detect the problem and degrade gracefully to regular picturefill behavior. Otherwise, pickyfill is fully supported in Android.
Use picturefill the same way you would without pickyfill. The only changes will be to load
picturefill.js and to give your site/page an HTML5 Appcache manifest.
<html manifest="/manifest.appcache"> <head> ... <script src="/assets/js/matchmedia.js"></script> <script src="/assets/js/picturefill.js"></script> <script src="/assets/js/pickyfill.js"></script> ...
Although it would be better to minify and concatenate the JS files, the above code is for clarity.
Because pickyfill will only cache images that are actually displayed, it is possible for a user to visit the site, then visit the site again while offline, resize their browser, and end up with a broken image (because the image that is required at the new browser size was never downloaded before and therefore has not been cached). For this reason, it is important to have an appropriate small
FALLBACK image in your offline Appcache.
FALLBACK: imgs/ imgs/fallback.png
Shibboleth authentication via simpleSAMLphp broke on one of our apps when the system clock on the Shibboleth server drifted more than a minute into the future.
I am not the administrator of the Shibboleth server. But it occured to me that even with no shell access to the server, I could detect clock drift by examining the time stamp in the HTTP headers from the server.
And Patrick Meenan’s awesome WebPagetest site has an API that allows for the automation of performance testing of web sites with actual devices.
Allow me to smash these two great things together and present my edu mobile web performance Reese’s peanut butter cup.
I screen-scrape Olsen’s list and have a Node.js script that instructs WebPagetest to load and reload each of the sites on an iPhone, measuring the performance.
The results end up in a publicly available Google spreadsheet.
I expect to update the spreadsheet at least once a day.
I am excited about this data to an unhealthy degree. More to come, as I look more closely at various sites to see what they’re doing really well and not-so-well.
What really set me down this path was discovering Mobitest. Their results will assign a decile for your site’s load time. Being unhealthily competitive, I of course wanted to see m.ucsf.edu in the top tier with the adorable “90th Percentile” indicator.
I went down a load of dead ends in the process. Here’s what worked for me (and a few things that didn’t).
If you are on a touchscreen device, you can see it in action at UCSF Mobile.
And here’s the obligatory video demo:
I recently had to add a map to the UCSF Mobile Maps site. To remember how to do it, I had to watch a series of videos I made after I did it the first time around. (Sorry, I didn’t have my presentation mojo happening, but the information is good.)
I could have documented what I did in a more conventional way, but here’s the thing: I wouldn’t have been as meticulous. I would have included what I felt was the minimum that I would need to repeat the task. I would have glossed over things that were obvious to me because they were fresh on my mind. But after months of not thinking about the task, going back to that sort of documentation would have meant that there would be substantial gaps.
By creating something that would live on YouTube, where I knew others would see it, I was motivated to be complete in my documentation.
It’s tempting to focus on things that aren’t very important when it comes to documentation. People like to create documentation templates, for example, wherein they try to have a section for every imaginable category of content that might be in a document. Most of the time, there will be tons of unused sections when the real documentation is written. And then the technique backfires. If someone is given a documentation template that is not even wrong, then there’s an excellent chance that they will just fill out the minimum that they can get away with and move on to a more meaningful task. In other words, they will just create bad documentation.
Of course, that’s not to say something ridiculous, like “All documentation templates are bad.” A template can be great for someone who doesn’t know where to begin or otherwise wants direction, especially if the template is reasonably succinct. (That said, an example of stellar documentation is usually better than a template.)
And there’s a lot that can be improved in my YouTube documentation. (A video is a terrible way to pass along a shell script.)
But, ultimately, nothing beats a documentation method and format that excites the documenter.
A few weeks back, on a developer conference call for the Mobile Web Framework, Ike Lin from UCSD presented data about the network payload size of various higher education mobile web site home pages.
UCSF’s mobile web site did pretty well in the comparison, but there was room for improvement.
Those improvements have since been implemented. This short video discusses what tools were used and how effective each was.
Tools discussed are: