What happens if Google offloads some compute calculations to every phone with Gmail
Billions of phones have the gmail app installed today. In the last 10 years compute power available on the phone has exponentially grown. Why doesn’t GMail / any other big consumer app business move certain computations to the phone especially computation that they do on the server side. This isn’t a new idea - block chains, protein folding, etc all rely on computing power from consumers. It would be cool to see users get back actual $$$ for allowing their phones to be used. For eg. Google could make YouTube premium free for users that allow sharing computing power of their phone. TC 620k Staff SDE
Sounds like you just saw the show Silicon Valley and got inspired
Battery life would suck. Nobody wants that.
Depends on how much compute. Google also owns the phone hardware for Pixels and has a lot of influence with other manufacturers. Similar for Apple.
can’t believe i have to explain this to a staff eng- it is just simply not worth it. we have massive warehouses full of the fastest computers we can make, why the hell would we use billions of shitty phones scattered across the globe?
Fighting spam is a good use case, go read about how the block chain was being used before Bitcoin. 🙂
+1 to google. This is a ridiculous idea. Why get people all mad for a negligible speed up? The thing now is using massive DCs to make ML models that phones can run on specialized hardware, not using phones to do the training.
An iPhone can pull 12 watts for an hour until the battery dies. A standard US wall socket can supply 1800 watts continuous. One datacenter can pull over 100 megawatts. You do the math.
An iPhone has more computing power than the computer used on the Apollo 11. You do the math.
Have you done the math yourself? How many mm^2 of wafer goes into iPhones vs datacenters? Actually the hardware is mostly a distraction here and watts are a better measure if you want to simplify the calculation.
This system is only good for very specific workload: — a task with high parallelism — very small input data footprint and output — highly computation intensive — insensitive to time of completion And I think it is overly simplfied, I pretty sure mobile processor instruction set is different from server-oriented cpu/gpus/fpga/soc/asic and other fancy words. Mobile processors and operating systems have their specialization for a reason. A slow/fast running application running a background workload will prevent app from saving resources (battery) and many users could become very upset. So such app won't be able to calculate that much before wasting allowed budget. I am curious if I've missed more/major/bigger points.
You are smart. I like the answer.
Would you risk people deleting gmail app vs small benefits from compute tasks moved to phone? How would you like your gmail app doing random compute all the time? Edge computing is cool but needs to be done for the specific use cases where the user is getting a direct benefit from the compute on his device. I bet Google has hundreds of thousands of over provisioned and under utilized machines in data centers sitting idle. Those can be better fit for such use cases.
Can you provide an example of a specific compute task you have in mind?
Well personally id immediately uninstall any app that tries this
Microsoft doesn’t need to worry. I don’t think anyone installs any Microsoft apps on their phone.
The typical use cases where this is useful are leaning towards preserving privacy. Also, the device architecture differences make it absolutely irritating to be able to deploy edge computing products while keeping the user experience largely consistent.
Did you just finish watching Silicon Valley?
Oh yeah, it’s been a while since I saw that show