DreamLab

~ the eulogy ~

DreamLab launched on 12 October 2015 with a proposition that sounds like a scam and was not one. Install it, leave your phone on the charger overnight as you were going to anyway, and while you slept the handset pulled down small packets of research calculation, worked through them, and sent the answers back. Imperial College London supplied the science. The Vodafone Foundation paid for the plumbing and the marketing, and the Garvan Institute of Medical Research in Sydney had been in it from the beginning.

It was discontinued on 2 April 2025, nine and a half years later. No reason was published. The final Android build shipped on 25 March 2025, eight days before the switch-off, which is not how an app behaves when a shutdown has been planned for a long time.

I work on inference at the edge now, so I spend my days on the same question DreamLab was asking, only from the other side of it: how much real computation you can get out of a small device that nobody is watching. My answer is usually thermal limits and a battery budget. DreamLab's answer was to wait until the device was plugged in and forget the battery entirely.

~ the dossier ~

~ how it worked ~

The pattern goes back to SETI@home and the screensaver-era projects: cut a large problem into work units and hand them out to whoever volunteers. DreamLab's contribution was to notice that the largest idle computer on the planet had stopped being the office desktop and become the phone on the nightstand. Roughly 87 MB of app, with the computation happening only while the charger was in.

Imperial's pitch put a number on it that was easy to repeat: one hundred thousand phones running DreamLab would match a year of output from every research computer at the college in about three months. The projects it actually ran were not filler either. Billions of combinations of existing drugs and food-derived molecules were screened for cancer-fighting properties, the Garvan team used it to find a new way of clustering cancers together, COVID and then Long COVID work came through it, and with the Grantham Institute it modeled tropical cyclones into what were described as the most detailed open cyclone maps available.

~ the next chapter ~

Volunteer computing keeps dying of the same thing, which is not lack of volunteers. GPUs in datacenters got cheap enough that the coordination cost of a million strangers' phones stopped being worth paying, and the research groups that used to beg for cycles now rent them by the hour. What the phone projects had instead was the part no cloud invoice buys: half a million people who could name a disease their charger was working on.

Vodafone's own pages about it are written in the past tense now, under a heading about a dream to help fight disease. The app stores have nothing.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »