Some calculations on whether Big Brother is recording every cell phone conversation

The idea that the U.S. government records all domestic communications has been in the news recently. That seemed like an incredible claim, so I decided to do the math.

According to this random site I found with a google search, there are about 300 million cell phones in the U.S. If each of them talks on average for about 500 minutes per month (a not-so-random site), and all calls are U.S. to U.S., that works out to 300,000,000 * 500 / 2 minutes per month, or 75,000,000,000 minutes per month, or about 1.25 billion hours per month. Of course, that doesn’t count non-cellular calls.

If it takes 10 megabytes to store an hour of phone call, that’s 12.5 billion megabytes per month, or 12,500 terabytes. Let’s say that you can buy terabyte drives for $50 in bulk. That’s $625,000 per month in hard drive expenditures. That’s less than five hours of flight time for a B2 bomber. There’s the question of servers, a data center, networking, etc. If that increases the cost by 20x, then we’re up to $12.5 million per month, or about $150 million per year, a rounding error in the federal budget.

Looking at it another way, one estimate (from a couple years ago) is that Google has about 1 million servers. That means that google could easily manage this data flow. Tack on speech recognition software for transcripts, and this kind of surveillance seems not just possible, but pretty straightforward and fairly likely.

Breaking out my tinfoil hat now…

1 thought on “Some calculations on whether Big Brother is recording every cell phone conversation

  1. gcanyon Post author

    Adding a note about speech recognition. It was possible to do reasonable continuous (real time) speech recognition in the late 90s. In some ways, computing power is more than 100 times as efficient now as it was then. If that translates to speech recognition, then it would take less than 20,000 computers to analyze *all* the speech in the phone calls in real time. In text form at 120 words per minute average, that comes to about 50 terabytes of text to store and index.

    Searching that database after the fact to find correlations and trigger phrases would be easy.

    Reply

Leave a comment