Is this the bridge too far, or will his supporters power through and find a way to believe this?

I skimmed the paper and wouldn’t take too much stock in it. One glaring issue I see is that the “control” for their analysis is another cell network that is absolutely tiny by comparison and services a completely different population from AT&T, so they’re not really isolating other variables, they’re cherry-picking data and over-indexing on it.

The other thing that immediately jumped out at me is the format-- it sure looks an awful lot like an academic thesis, and the authors are a professor and a student who started this work during their undergrad. There’s surely no way a respectable journal would publish this document in its current iteration, so why is it in the news? Weird.

The webpage he hosted was a copy of his own blog post explaining the hack. It just about fit into the 20KB of available flash storage.

We can infer that on every request, the whole static page needs to be spooled out of flash onto RAM (in chunks no larger than 3k), then sent out over Ethernet.

That’s an awful lot of work for the chip. I’m not surprised at all that it errors out under heavy load. The request queue probably grows until it collides with the buffer that bucket brigades the web page to the network.

I’m afraid to look up what optimizations were necessary to get that level of performance. It’s damned impressive work.