I suggest you start by simply comparing the cost of dedicated hosting-provided hardware. This is the middle ground between owning your own stuff and something like AWS.
Pricing out the cost of owning your own servers can be tricky since you need to account for a number of factors:
* Initial cost of hardware - don't forget to include things like network switches, remote console devices, etc. - you may also want to buy something like a remotely accessible power switch so you can forcibly reboot machines
* Rack space costs
* Bandwidth costs
* Sysadmin time to manage the hardware (you already need a sysadmin to manage things like backups, logging, etc.)
* On-site remote hands if your rack host is not physically near your sysadmin
But regardless, I'm pretty sure that _anything_ ends up being cheaper than AWS once you get beyond a fairly small number of servers. The big advantages of AWS are integration with other services (storage, load balancing, etc.) and the ability to spin many machines up and down very quickly.
However, if you can tolerate waiting a day or so to provision new servers, dedicated hosting-provided hardware is almost certainly going to be quite a bit cheaper.
You could also compare this to something like Linode. Again, you lose the ability to provision new machines almost instantly, but they are pretty quick to bring up new machines (minutes, not hours or days). They also have a number of nice integrated services like backups.
Finally, you can also do some sort of hybrid setup where you have core stable services on dedicated machines but still spin up AWS instances to deal with bursty computation needs.
Of course, raw hosting costs are only part of the calculation. You have to include sysadmin time and salary.
Of course, if you don't need permanent storage (and your needs are bursty) than using EC2 is probably a huge win.
My takeaway, the economics of a lifetime storage offering are very difficult to achieve. Glacier changes the rules by altering the SLA for data access to make the money work out.
I work in a very different area, but my experience with ec2 are you should drop what you're doing and immediately price out leaving. Roughly: for big hadoop tasks, ec2 is a horrid environment. The boxes are slow, flaky, and shitty; the network is slow and super shitty; and everything about the experience is ass. Oh, and the emr admins are assholes who push broken code; code that couldn't even have run once. These assholes, after issuing us a hotfix to a boot action, proceeded to change the path to the hotfix (without mentioning it to us), breaking our production clusters twice in in 30 hours. I guess I hold a grudge but I like sleeping. The point is emr and perhaps ec2 are complete amateur hour.
Quoting myself earlier:
from a former employer who wants to stay anonymous, ec2 bill per month: $97k
peak, softlayer: $26k. 1/3 the cost, 2x the performance.
And let's not mention emr; a $15k/mo cluster in softlayer did something like
5x the performance of an emr cluster that was costing (and running!) $2k/day.
This was over 5 years ago. I am sure things would be even cheaper now.
The 244gb cr1.8xlarge (16 cpus, 240gb ssd) amortized over 3 years costs $659 / month. Softlayer would charge quite a bit more for similar specs ( order of magnitude more).
When you factor redundacy I need to buy 200tb, factor in other server infrastructure with the cost of a 4TB drive running around 180 and I can spend 15K to buy 200TB of storage.
Put another way, we can own twice the storage capacity (or 1x mirrored) for roughly 2 months of S3 storage.
You're also completely ignoring the costs of hosting this stuff, the computers they're in, redundant network switches, failed disks over time, and administration costs to rack and stack failed drives plus all the risk associated with your redundancy going wrong. You're also assuming your hard drives will be able to handle the performance needs. Not to mention this is one site - no DR.
So, Amazon S3 at $7.5k a month for 100 TB is $90k annually with RIDICULOUSLY high redundancy and known performance. Reduce it to 99.99% availability (about what your setup with dual drives might be if you run it well) and you're at $60k annually.
You really are going to host fast, reliable 100 TB for under $60k? Betting your business on it?
It's far from simple though: you need to find the people who can build such systems and also be able to afford the capex. And then there's the risks of running your own system and opportunity cost.
It's really unclear we would have been able to go this far in the first place without AWS.
It's really unclear we would have been able to go this far in the first place without AWS.
You could have drastically cut your burn on that last $550K by hiring a sysadmin and bringing it all out of AWS (maybe keeping S3). This would have likely also involved retooling on the app side, but in the long run it would have given you more time for positioning/promotion/etc. All in all, 4th Gen is way late for this.
P.S. Number of photos is basically irrelevant, were you actually using market-fit metrics for infrastructural decision support?
pricing out servers and researching / working with smaller infrastructure providers who actually care about you and your account (imagine that) is real work, which is why so many people avoid it these days.
it's much easier to click a few buttons and "choose" machine type x in the offerings that amazon has designed to maximize their own profit.
computers are hard, let's go shopping.
1) Switch to other cloud providers, I'm not aware of any very cheap ones.
2) Switch to other hostings, which can be 1/10 as cheap as AWS, but then you have to deal with the failures.
3) Colocate and build your own servers/storage. It's only cheaper if your bill is much more than a 24/7 sysadmin(s) salary.
I suspect DropBox don't give a shit about server costs, they are swimming in money. But I bet having their own data centers will save them money.
If you are a photo backup company, storage is (or at least should be) part of your core.
It seems like they tried to take the heroku route, repackage aws services with premium software and sell it for a premium. But forgot to charge the premium end user price.
Total Salaries (between 6 people) $1,168,710.45
Total Payroll $1,298,819.67
Total Personnel Costs $1,411,513.53
That's about 10x their hosting expenses. Even if they had free hosting it seems like it would only have kept them going for another 2-3 months.
When you get millions in funding its easy to piss away on the highest tier services. When you start from the ground up, every cent is managed.
As someone who has worked in this space, I can tell you that the skills simply do not exist. They don't hire for them, because they all ("all") use outsourced hosting and deployment frameworks that insulate them from these things.
Plus, you don't have that all-important "Cloud" buzzword for them to throw around!
I'm 20 myself, and while I'm not the hottest shit around, I have been working at a bank for the past two years. I've been exposed to some really smart people with extremely challenging problems, way more so than creating a pretty service in a few weeks and getting crazy amounts of funding for a business that's most likely going to fail because we're just that, kids.
Before coming into my role, I truly did think I was the best because I was able to create pretty websites with whatever is the latest JS library everyone is fawning over. That illusion was destroyed immediately when you start designing and working on infrastructures that have to handle millions of payments a minute, working with multiple vendors, business units, testing units, and various other departments you need to familiarize yourself with.
Unless going custom allows you to do something that you couldn't otherwise do (cases which are rare, but existent), it's just optimizing margin. And you can't margin your way to success. (To failure yes, to success no.)
As noted above, their hosting was a big number, but it wasn't a dominant cost. It's like Amdahl's law of money, it doesn't matter how little you spend in one area if that area isn't dominating your costs.
And all else being equal, it's usually the wrong choice; the best way to grow the bottom line is to grow the top line. If you can spend X weeks cutting costs by Y$, or X weeks increasing revenue by Y$, grow the top line. It's fuel, it gives you options.
The thing I miss the most about doing it our selves was the raw fun of it, and how efficient it felt. But in retrospect it was a lot of time spent on things that ultimately didn't change the outcome.
That is insane to me. I make less than 80k a year, in D.C., with a mortgage and two kids while I build on the side. A cursory glance tells me most founders aren't under such strict cost regimes. If and when I get funding and go full time I would take the absolute minimum to keep us fed and watered (~60k). How a founder could do otherwise to me is unconscionable.
I didn't say pay yourself nothing - pay yourself the minimum amount you have to in order to keep your life from impeding your work. If that means you need luxuries to be comfortable, then your lunch will get eaten by someone who is more spartan.
Did you have intermediate milestones before reaching the "economies of scale" one?
Would you mind sharing the P&L statement and/or pitch deck that the verge used in its reporting? The verge's article seems confused, and I think one of the best gifts you could make to the HN community is to teach us from this outcome with actual source documents.