Comment by mystified5016 | Hacker News Reader

Comment by mystified5016 | Hacker News Reader

Parent Full thread

At the startup I work at, we have an AWS instance that only runs our GitLab server. The damn thing runs at 10-15% CPU nearly constantly (because GitLab's founding assumption is that everyone wants to scale to ten million global users) so our spend racks up to $70/month every months.

And yes, that's absolute peanuts for any business. But in my view, spending almost $1k annually for a GitLab server for a team of ten is ridiculous.

We could accomplish exactly the same thing on-prem with hardware we already own. It would take a couple of engineer-hours per year. As long as it's under 40 hours of maintenance per year, we come out day ahead. And over the last two years, I've had to spend a total of maybe 10 hours with hands on this box.

I just don't get the thinking that leads businesses to put unnecessary crap in the cloud. Just save your money, on-prem is cheaper for pretty much everyone not doing a saas or less than a few hundred employees

Your estimate is off by an order of magnitude unless you value developer time at $21 an hour. I pay union electricians 6x more than that to work for me.

> $70 a month times 12 is $840

> As long as it’s under 40 hours of maintenance (??)

I don't know where the person in question is, but those numbers seems low if you assume zero hours of maintenance for the cloud setup, sure.

The thing is, you're not likely to spend amortised more than minutes a year maintaining the on-prem hardware (or you could rent managed servers and spend nothing on it), and as they pointed out they already spend hours maintaing the EC2 instance.

I wouldn't typically recommend on prem hardware for just a single server - if you already have a rack and a proper setup, sure - otherwise just rent a cheap server somewhere, and ensure you have backups.

I often starts that way, yes. And then you end up accreting more and more.

I usually tell people that if they don't want something on-prem, then at least pick a cheaper provider.

But as long as it's a VM and you don't build in reliance on other AWS services, at least you can move off it as you scale. The real problem comes when you start depending on a single cloud.

Not only do you lose the ability to do an easy move, but you also lose almost all negotiating leverage as your bill increases.

My mental model of startups is that they're free of the bureaucracy that would cause you to whine in comments versus Just Fucking Do It. So, what's the impediment to using your time to test your "10 hours" theory?

Also, my experience with GitLab isn't that the thing is hurp-durping because of scaling to ten million, it's that ruby gonna ruby

Reply on news.ycombinator.com