Hacker News Reader

Comment by gruez | Hacker News Reader

residential customers have really poor upload speeds so how many customers would you need to have to replace 1 caching server collocated at an ISP? short of having FTTH, most customers would have < 10Mb/s up. a dedicated server can have multiple 10Gb/s links. you'd be looking at hundreds or thousands of customers per caching box. the "nominal discount" you're offering would have to be tiny for it to make sense financially. so much so that most customers wouldn't even bother.

View on HN

If the nominal discount was $1/month, most customers with an unmetered connection would take it because it's effectively free money. From the company's perspective it could even be revenue positive by raising the price $1 and offering it back as the discount.

You're also not considering the power law distribution. If nineteen customers have 1Mbps uploads and one has symmetric gigabit, the average is >50Mbps. And the discount could be per-byte -- set it at half whatever the cost of central servers would be.

Watch how quick those unmetered connections become metered once this takes off :P Companies doing the distribution value control, this would be quite difficult to administer. People put up with flakiness only because they're getting something they shouldn't be, or because it's free anyways. Once people start paying, it becomes a very different set of expectations.

Put on your product hat: as Netflix, you want to control every aspect of the experience to ensure its absolutely seamless. When you outsource distribution to a huge set of third parties that becomes much harder to achieve.

My solution to this for a video streaming service is to have a master network that you control that feeds the data needed to cover the next 30 seconds of video playback, with future segments coming in via torrents. If any segment is still downloading once it hits the 30 second threshold the connection is dropped and that segment is downloaded from your more than capable network.

Connections that are known to be slow can seed just the ends of files with faster customers seeding the entire file.

I imagine the maths on this has been done and the cost of bandwidth is much lower than the cost of engineering the solution though.

Update the protocol to securely track who uploaded what, to whom, and you've got a market just like private solar panel owners and the energy grid. Pay the 1mbps and gigabit civilians proportional to how much they share.

It's entertaining to expand the idea beyond bandwidth...what about CPU/gpu time on my powerful consumer desktop sitting idle/off all day? Or storage space? The computer would need to be on to seed... Would people pay to also use the idle compute capacity?

Security is a problem, but it could be used to segment customers and providers. (Big corporations bidding for AWS capacity, tiny personal sites hosted on a CDN of cheap consumer hardware).

Tagline it "Uber for the cloud"!

Filecoin[1] is doing that for storage+bandwidth and Golem is pursuing the CPU/GPU time idea.

[1]: https://filecoin.io [2]: https://golem.network

>most customers with an unmetered connection would take it because it's effectively free money

not really, because have to pay for the electricity costs of leaving your computer on 24/7. assuming a modest 50W power consumption for a PC, that works out to 36KWh per month, or $3.6 at 10 cents /Kwh. That alone wipes away your $1/month you're saving. also, that's a third of the monthly plan cost, how much of a discount can netflix really offer here?

as for hard disk space, I'll grant it's "free" space (as in the consumer wasn't going to use it anyways), but running consumer drives 24/7 with frequent seeks can't be good for them.

There is no need to have the consumer PC running 24/7. Also there is no need to have the same uplink speed as downlink speed. Consumers on average just need to reach a seed ratio of 1, while downloading and consuming the content and then possibly doing other stuff on their devices until the content is deleted. Actually, if they only reach 0.3 on average, then still 30% of the server infrastructure is saved, which is already a lot.

How many people turn off their computer regularly? I suppose people with laptops probably do. I leave my computer on 24/7/365.

Most people that are not in IT.

"Approximately 50% of computers are in On mode for 4 to 5 hours or less."

  https://ees.lbl.gov/sites/all/files/computers_lbnl_report_v4.pdf

I wouldn't have guessed.

I have unlimited 300 Mb/s upload and it's nothing unusual in my country (Kazakhstan). I guess it depends on country.

I gave 200MB/s down and 10MB/s up. I suspect this is more normal.

Yes, it depends on country. But I also suspect the total upload bandwidth for typical users in typical locations simply isn't there to make the idea work at scale.

What does the ratio of customers to caching boxes have to do with the size of the proposed discount?

because if you're offering customers $1/month discount to use their computers as cache boxes, but dedicated 1 caching box is equivalent to 500 customers in terms of capacity, then if the amortized cost of 1 caching box is less than $500/month, it's cheaper to go with the dedicated cache box than to pay customers to run cache boxes.

(all numbers are hypothetical)

Reply on news.ycombinator.com