Live data from Hacker News

Google's Subsea Fiber Optics

cloud.google.com

11–20 of 103 posts

Re: Google's Subsea Fiber Optics

#11
post #6

I make 340Tbit/sec about 1.1x10^11 GiByte/month. GCP premium tier networking is priced at $0.08/GB, so at 80% load that cable would, very naively, have the potential to bring in $7B/month in revenue. I'm sure they only take in a fraction of that, and their costs are substantial. But even so... cloud bandwidth is overpriced.

Also 80% utilization seems just ridiculously high to me but maybe at goog volume it’s doable

Re: Google's Subsea Fiber Optics

#12
post #6

I make 340Tbit/sec about 1.1x10^11 GiByte/month. GCP premium tier networking is priced at $0.08/GB, so at 80% load that cable would, very naively, have the potential to bring in $7B/month in revenue. I'm sure they only take in a fraction of that, and their costs are substantial. But even so... cloud bandwidth is overpriced.

Yup, but it will continue to be absurdly overpriced because the CapEx is massive and governments are totally okay with oligopoly.

Re: Google's Subsea Fiber Optics

#13
post #6

I make 340Tbit/sec about 1.1x10^11 GiByte/month. GCP premium tier networking is priced at $0.08/GB, so at 80% load that cable would, very naively, have the potential to bring in $7B/month in revenue. I'm sure they only take in a fraction of that, and their costs are substantial. But even so... cloud bandwidth is overpriced.

Also 80% utilization seems just ridiculously high to me but maybe at goog volume it’s doable

I remember seeing a cloudflare post about AWS bandwidth pricing where they estimated something like 20% utilization? I don’t remember where though but I think they can approximate pretty well.

Re: Google's Subsea Fiber Optics

#14
post #6

I make 340Tbit/sec about 1.1x10^11 GiByte/month. GCP premium tier networking is priced at $0.08/GB, so at 80% load that cable would, very naively, have the potential to bring in $7B/month in revenue. I'm sure they only take in a fraction of that, and their costs are substantial. But even so... cloud bandwidth is overpriced.

You’re thinking about average throughput, while these cables need to be provisioned for max throughput, which can be completely different.

Having said that, cloud bandwidth is indeed overpriced; but at the same time, given that Google Cloud is still burning money, can it perhaps be argued that bandwidth is one of the money makers that allow for other services to be free?

I recall that from the old webhosting days, this was already a common tactic of the providers: lure people in with cheap servers, sometimes even at a loss, and earn money back with bandwidth.

Re: Google's Subsea Fiber Optics

#15
post #5

I wonder how long it will be before we see the first hollow-core fiber subsea cables. They are 50% faster, and tests from the last year or two have seen record low signal losses. https://www.laserfocusworld.com/fiber-optics/article/1419605... https://www.ofsoptics.com/wp-content/uploads/Hollow-Core-Fib...

> They are 50% faster

No, pretty sure light still travels through them at C. What they can do is carry more data, largely by having lower error rates.

Re: Google's Subsea Fiber Optics

#16

Earlier quoted context omitted.

People happy to share their work. Is that some kind of problem?

The video doesn't look like that, it looks more like everyone was told "you have to smile more!!"

This seems a super-cynical take. Some people are smiley and happy naturally. You might pick them to be in a video.

Re: Google's Subsea Fiber Optics

#18
post #6

I make 340Tbit/sec about 1.1x10^11 GiByte/month. GCP premium tier networking is priced at $0.08/GB, so at 80% load that cable would, very naively, have the potential to bring in $7B/month in revenue. I'm sure they only take in a fraction of that, and their costs are substantial. But even so... cloud bandwidth is overpriced.

Also 80% utilization seems just ridiculously high to me but maybe at goog volume it’s doable

Google regularly runs these at 100%. According to the B4 paper:

"""These features allow many B4 links to run at near 100% utilization and all links to average 70% utilization over long time periods, corresponding to 2-3x efficiency improvements relative to standard practice"""

Post reply on HN