Bug In Samsung S3 Grabs Too Many Images, Ups Data Use

First time accepted submitter Emmanuel Cecchet writes "Researchers of the BenchLab project at UMass Amherst have discovered a bug in the browser of the Samsung S3. If you browse a Web page that has multiple versions of the same image (for mobile, tablet, desktop, etc...) like most Wikipedia pages for example, instead of downloading one image at the right resolution, the phone will download all versions of it. A page that should be less than 100K becomes multiple MB! It looks like a bug in the implementation of the srcset HTML tag, but all the details are in the paper to be presented at the IWQoS conference next week. So far Samsung didn't acknowledge the problem though it seems to affect all S3 phones. You'd better have an unlimited data plan if you browse Wikipedia on an S3!"

The relevant quote with fault highlighed

[tuppe666]

The Samsung S3 browser bug
======================
When comparing our results on the different devices and networks for our Wikipedia trace, we noticed significantly higher latencies for our Samsung S3 smartphone on both Wifi and 3G. We first looked at the number of HTTP requests per page and the size of the pages down loaded from the server. Our findings are illustrated on Fig. 13. The number of HTTP requests is always much higher for the Samsung S3 and the page sizes are much bigger. Note that the page size for Samsung S3 on 3G is sometimes very small as we only account for successfully transferred bytes and not expected object sizes. On a successful page load, the page sizes should be the same on both networks. Fig. 14 gives an insight into the cause of the problem. By
looking at the recorded HTML page source, we saw that Wikipedia pages use srcset HTML tags that indicate a list of images to pick from depending on the resolution and magnification needed by the device. It turns out that the S3 browser has a bug and systematically downloads all images in a srcset instead of picking only the one it needs (left most red circles on Fig. 14 show 3 different versions of the same image being downloaded). This can result in a massive amount of extra data download.

The Wikipedia page dedicated to the Internet Explorer browser that typically requires 600KB of data download jumped to 2.1MB on the S3. This bug significantly affects the Wikipedia performance on 3G were these massive number of requests for image downloads overwhelmed the network and ended up timing out rendering an incomplete page. This can be seen on Fig. 14 where a large number of requests are blocked for very long amount of time and many of them fail with a ‘NO RESPONSE’ HTTP error code. Note that we were able to reproduce these results with the latest Android 4.2.2 for the S3 GT-I9300(international version of the phone). The issue was also reproduced with an S3 SGH-I747 which is the AT&T US version of the phone. We believe that this problem affects all S3 versions and have contacted Samsung to report the issue.Having a database with results from other devices helped us to quickly locate the origin of the problem and detect this previously undiscovered bug. Based on this experience, a possible direction for future work is to design tools that automatically analyze and report anomalies by comparing
experience reports between devices/networks for the same trace.

Re:Moot

[Anonymous Coward]

The vast majority of people use the stock browser, and defaults in general. Not everyone is a geek.

http://www.androidpolice.com/2013/04/03/according-to-net-applications-stock-android-browser-usage-is-still-way-out-in-front-of-chrome/

Bad Data

[tuppe666]

The vast majority of people use the stock browser, and defaults in general. Not everyone is a geek.

http://www.androidpolice.com/2013/04/03/according-to-net-applications-stock-android-browser-usage-is-still-way-out-in-front-of-chrome/

It does not change your point of your comment but netmarketshare http://marketshare.hitslink.com/browser-market-share.aspx?qprid=0&qpcustomd=1 [hitslink.com] where the data comes from, has something wrong with the way records data, especially with mobile usage. Its often quoted on Apple sites due to its heavy bias towards Apple(that does not reflect real world use). They have heavily massage figures, and they do not match those of independent larger sources. Here is statcounter http://gs.statcounter.com/#mobile_browser-ww-monthly-201205-201305 [statcounter.com] (Again it does not dispute your point but the source data)

Re:Nothing to do with Linux

[Nerdfest]

With Apple, you can install other 'browsers', but they're really just skins for the internal webkit engine, and they do not integrate fully with the OS.

Re:Company reactions to major problems.

[Emmanuel Cecchet]

Actually trying to report the bug to Samsung was quite hard. First there is no place to report such bug in the first place. The place that seemed the most appropriate was tech support but it showed that Samsung is a hardware and not a software company. The tech support can just handle hardware issues with the phone or basic user issues using the phone. When we submitted our bug report to them they were at a complete loss and didn't know what to do with it. The office of the CEO message was kind of a last resort measure but once again the supposedly R&D team that reviewed the issue dismissed it saying it was just an Android problem and didn't investigate further. To really attract their attention, we should probably have posted on their Facebook page but maybe this /. post will incite them to look into the issue again. The conspirationist will see a collusion between Samsung and carriers trying to squeeze more money from users by inflating their data usage. The engineer will just see a subtle bug that is not easy to catch by QA unless you can compare your device behavior with other devices and automatically detect such anomalies.

Re:This is a big deal..

[neonmonk]

As others have noted, Wikipedia is pretty much the only website that has even implemented src-set. This is not a big problem. This is a very minor problem.Maybe if the whole world was using src-set then it would be a big problem, but they're not, and won't be for a long time seeing as none of the big 4 browsers have implemented it.

Re:This is a big deal..

[_xeno_]

When considering 500MB is the usual data cap this is a problem, with the amount of data slurped up by the likes of Facebook, this must push useage up pretty high if loading a wikipedia page is taking over 2MB of data.

Not really, because Wikipedia is basically a worst-case scenario. To show you what I mean, here's the first <img> tag off Wikipedia's home page at present:

<img alt="The Tichborne Claimant" src="//upload.wikimedia.org/wikipedia/en/thumb/3/37/TichborneClaimantSketch_cropped.jpg/100px-TichborneClaimantSketch_cropped.jpg" width="100" height="137" srcset="//upload.wikimedia.org/wikipedia/en/thumb/3/37/TichborneClaimantSketch_cropped.jpg/150px-TichborneClaimantSketch_cropped.jpg 1.5x, //upload.wikimedia.org/wikipedia/en/thumb/3/37/TichborneClaimantSketch_cropped.jpg/200px-TichborneClaimantSketch_cropped.jpg 2x" />

The bug appears to be that it loads all three images specified - the 100px (from src), and the 150px and 200px (from srcset) versions. But that's because Wikipedia not only uses srcset, it provides three different resolutions: a default (100px), and two "high DPI" versions (1.5x and 2x). Most other websites don't even use srcset at all - because no other browser even supports it. Not Firefox, not Chrome, not even Safari despite srcset being an Apple creation.

Facebook doesn't use srcset, so it won't trigger this bug.

In fact, I don't know of any website that does use srcset other than Wikipedia. Google doesn't. Twitter doesn't. Facebook doesn't. Slashdot doesn't. (Nor does CNN, Fox News, the BBC, Yahoo, Flickr, Tumblr, or Amazon.com.)

It's basically a bug that will only trigger on Wikipedia, so no, it's not really a big deal because unless you spend a lot of time on Wikipedia, you'll almost never trigger it.

It's still a bug that should be fixed, but I'd be hard-pressed to call it a "big deal," solely because about the only way you'll trigger it presently is on Wikipedia.