Comment by Aissen
2 years ago
The cloud tax is crazy (especially bandwidth). Pretty sure Signal has reached the scale where they would be cheaper by building their infra, maybe starting with the most expensive (storage + bandwidth), and then doing others.
SMS is (unfortunately) core to the product, so I'm not certain how they could make it cheaper, while retaining the same properties (user+pass registration would be a nightmare for spam and change the UX).
Is it not that they’re buying something resistance as well by buying from large infra providers if big adversaries like state actors start pushing hard?
I also think SMS and phone numbers are core, but they must provide a way to communicate without use of phone numbers being kept completely separate from phone numbers even when registration is needed using phone numbers.
The usernames are coming (alpha was announced), but it won't reduce cost since the account is still phone number-based.
Anyone know much does it become worth it to build your own? They spend around $3-4m on storage and bandwidth
> millions upon millions of new people suddenly switched to Signal in January 2021 after WhatsApp updated their Terms of Service
From a footnote of the article. Maybe this is why they've stayed with "infinite scale, infinite costs" (commonly known as "cloud") so long? Surely at some point this is worth considering though, I would also be curious where that point lies
Virtually anyone, also when spending only 100 euros/month on server providers, can save a large percentage of costs by taking it in-house. There might be a gap where you need dedicated personnel and it's briefly cheaper to outsource before you grow and it inverts again, but generally if you've got a stable service then this is nearly always worth it
Maybe a hybrid, where new users onboard onto cloud and they buy hardware for expected loads (i.e. current users), would be the most cost effective. I wonder how hard that is to combine the two worlds, but anything that requires more than one server already has that sort of communication going on so there shouldn't be any real blockers. Maybe the two types of infra add costs/risks again and that's why one rarely sees this setup?
I know AWS - and I would guess the others too - discourage hybrid by setting the egress traffic costs to extreme levels
I found that with the bandwidth and storage that my company was using on the cloud, we could get ROI in under 2 months by building a server and running it in house. Now we've scaled up to a dozen servers but it's still just a handful of computers in a closet that saves us $50k/mo in cloud costs. It was dirt cheap to slap together and scale up incrementally.
Data centers cost billions. Signal, and pretty much everyone else who isn't already in the data center business, is far away from breakeven on that.
There are several steps between using AWS and building a datacentre.
- Using similar services from cheaper cloud providers
- Renting VMs
- Renting whole servers
- Renting rack space + power
- Renting larger spaces (many racks, or part or all of a whole floor)
The small ISP/phone/cable company I worked for in high school had a data center. Maybe 20 racks. It was pretty damn reliable (old-school phone infra techs knew how to make shit stay “online”). I guarantee it wasn’t above the single-digit millions to build, inflation adjusted.
2 replies →
You don't have to buy a data centre... You can lease space in an existing facility for almost nothing.