At one point it took often days for a dedicated server to be set up. We also didn't have such nice provisioning tools.
Now it just seems like cargo cult to use cloud providers as the only option. People just completely discount dedicated servers.
At one point it took often days for a dedicated server to be set up. We also didn't have such nice provisioning tools.
Now it just seems like cargo cult to use cloud providers as the only option. People just completely discount dedicated servers.
developer costs >> infrastructure costs
An AWS large server is around $500/year, which is about 1-2 developer hours (with taxes, overhead, etc) at the cost scales last time I priced this out. That's crazy expensive in the absolute, but if it saves a couple of hours, it makes sense.
PaaS providers cost even more. I've gone with those in the past, since it basically eliminated dev-ops. The major downside wasn't cost, so much as flexibility.
Dedicated servers start to make sense for:
- The very low end (e.g. personal use, or hosting something long-running for a small business)
- The very high end (e.g. once cloud costs start to hit hundred of thousands of dollars per year)
On the very high end, cloud providers will often cut a deal, though.
My problem with AWS, recently, has been reliability. Servers crash or have performance degradation a bit too often. That leads to developer costs, and might be what pushes me back to dedicated.
Anyone looking to hire an ML guy for 250-500$/hr flat fee, hourly rate, no taxes, no healthcare, cancel any time... get in touch!
For that money I'll gladly bark like a dog, walk on all fours and fetch your slippers.
In the US salaries are a lot higher so in that context I can see 300-500$/day be realistic.
Really surprised me… Always made me wonder if all that money was being spent well
You should see how much lawyers cost versus how much lawyers earn. The sticker price for SWEs isn't as well-advertised, but it's how businesses work.
In your case, you are not exchanging 2 hours of a developer for a higher bill. In my experience you will get a higher bill and 5 hours of a cloud expert, and probably a solution that for every change or problem, you will have to call this expert as no one know what to do.
> spend way more time with Lambda/Cloud formation/ECS
As you said, it's not your expertise. No one said AWS could be used well by non-experts. The point isn't the learning curve, but the terminal velocity once you get there.
Trying to use a thing that accomplishes the a similaroutcome but in a totally different way requiring a totally different skillset, it's not surprising that you found yourself needing someone who does have that skillset to assist.
A Linux server guy trying to manage Windows servers will probably need help from a Windows server guy. A car driver trying to fly a Cessna somewhere will probably need help from a pilot.
That expert may not know what to do either, or... will contradict earlier folks and say "you need to redo all of this pile..." and you'll be stuck.
I've no doubt there are some workloads that really do require a level of complexity that various cloud systems offer. Much of what I've seen doesn't require it, but once someone starts down that road, they're 'justified' in learning more and tying more stuff to the cloud provider's way of doing things.
I cut my teeth setting up servers back in the 90s, and that you can spin up 'full' servers (thinking various shades of VPS) in a few seconds is crazy great. But we're somehow now 'past' that... and we have mountains more complexity to consider.
The massive number of proprietary newer services are junk, with the exception of some of the ML services. The ML services are sometimes helpful, but I expect they'll quickly be obsoleted.
That's what we said at my first Kleiner company. All the time.
Pretty soon we had a $100k/mo server bill.
We were just webhosting. There was no need for any of it.
You should be thinking of dev time as in _investment_ not a cost. Invest developer time into high value activities like creating robust services you own. Not into low value activities like coding against a dozen different proprietary, Amazon owned APIs that can change, go away, and become more expensive any time they feel like it.
> developer costs >> infrastructure costs
What I've expected alot is the businesses are fundamentally struggling to find sysadmins therefore paying for infrastructure costs are there only option. And now more and more sysadmins are "AWS devops" who literally only know how to manage a AWS stack with many struggling even basic stuff such as managing to figure out how many resources they will need since more and more stuff are autoscaled.
They can hire devs but can't hire sys admins. Hell some of the time they can't even hire devs.
For comparison's sake, the net salary for a software developer in Latvia is roughly between 950 and 2800 euros per month: https://www.algas.lv/en/salaryinfo/information-technology/pr...
Given the above number, some calculations of the gross salary land at between 1350 and 4000 euros per month: https://mansjumis.lv/darba-algas-kalkulators
So essentially one of your developer days would represent weeks of work for a Latvian developer (depending on additional expenses/overhead), which I find curious.
Developer costs are still likely to be higher than infrastructure costs, but not nearly by as much, which might explain why I've seen a lot of on-prem deployments locally, and people shying away from going all in with cloud technologies.
I wonder whether these differences are more pronounced for other countries where developers can be comparatively less expensive (at least in relation to using the cloud and/or hardware), like India, Russia and elsewhere, how much the development culture differs. Also how many would opt for alternative providers like Hetzner, DigitalOcean, Vultr, Scaleway, OVH and so on...
For the use-cases I was involved in (image gallery hosting, 2-3Gbps traffic), costs looked like:
AWS: $400/mo for app servers, $40,000/mo for CDN bandwidth
Bare metal: $400/mo for app servers, $400/mo for a couple of caching reverse-proxies, each with an unmetered gigabit connection
When you’re dealing with anything more bandwidth-heavy than HTML+JS+CSS, having a handful of bare-metal Varnish instances is _orders of magnitude_ cheaper than the cloud…
(We have also looked at dedicated CDN providers like cloudflare, but we’re in that sweet spot where we’re too large for the publicly advertised plans, and the “call us for a custom contract” plans want a minimum spend around $3,000/mo :/ )