Comment by Beltalowda | Hacker News Reader

Comment by Beltalowda | Hacker News Reader

Parent Full thread

> simply put their workloads in the cloud and be done with it

I've worked all sorts of cloud environments for about ten years now, and with metal servers before that, and I'm not all that convinced that the cloud is saving all that much in terms of engineering effort. Or if it does, I haven't experienced it.

Sure, no one gets woken up at 4AM because your server's disks are failing, but instead you get woken up at 4AM because all your cloud native infrastructure is doing $funky_stuff. And yeah, no need to install and maintain servers any more, but now you need to write a whole bunch of code to interface with all the cloud stuff and maintain that. Lots of open positions for k8s engineers "dev-ops" engineers to make all of that work.

I run my stuff on Linode VPS's, which is "the cloud" as in "renting servers", but it's not "the cloud" as in all the "cloud native" stuff we have now. My experience is that sometimes the cloud makes things easy, and it makes things harder. On balance, it seems we just swapped one set of problems for another set of problems, and we haven't really gained all that much in the end.

I don't think Hey is running a "niche edge case"; it seems fairly standard stuff to me. That Tweet comes off as quite dismissive.

The tweet thread came off in the completely wrong way to me, and suggested that the person writing it was pretty ignorant about how software engineering actually works. Are you suggesting that everyone should be using [insert dev fad here] for all but a few special cases?

Have you considered that sometimes (really, almost all the time) people run code that is more than 2 years old? Have you considered the costs of things like serverless when you have a reasonably constant load?

I am a huge fan of cloud as a platform for time-shared server rentals. Cloud for "value added" products like serverless and whatever new database they cook up starts to get really funky and weird, in addition to being extremely expensive.

Wardley assumes serverless is the continuation of decades of cloud evolution.

He doesn't see it as a fad to use the cloud and serverless, but as the natural way of things, like you don't build your own power plant or highway.

Wardley seems to have a narrow perspective on things, then. Lots of large companies build their own roads and power plants. Hell, I put solar panels on my roof.

In fact, there are very few companies over a certain size that haven't built a road. This kind of overhead is generally expected once you have scale, because the alternatives are far more expensive. Datacenters are similar.

How much of all companies build their own roads?

Pretty much every company that builds an office building builds a road or two.

That is true, but only a tiiiiny fraction of all companies build office buildings.

If you're one of them, go get your datacenter.

Just to extend that analogy: real estate and construction companies tend to build a lot more roads than other companies. Even small construction companies have built a decently long road on some project, and the smallest have built a long driveway. Specialist road building companies that build a ton of roads exist, but all construction companies can build a good road.

By the same analogy, tech companies should not run from building server racks and/or datacenters. It may not be your company's core competency, but it should be one of your competencies to understand how to run a computer. It really doesn't take that much work to run 10 computers.

People keep using the power plant analogy as if somehow there isn’t a massive shift towards solar and batteries, which are literally building a power plant on your roof.

Well, you're literally not building and maintaining it on your own.

No one is really "building" anything when they run their server though; it's all built by hardware vendors, you get the software from software vendors, etc. You're just operating it.

I don't really like using analogies for this sort of thing as I think it muddles things more than it clarifies, and quickly leads to discussions about the analogy rather than the topic at hand, but if we must use one I'd say it's more analogous to a business deciding to purchase a company van or lorry. For a lot of businesses, this clearly isn't needed and renting one when you do is fine, but at some point it just becomes easier and more cost-effective to have your own (depending on what you do with it, nature of your business, etc.) in spite of the extra hassle of having to maintain, clean, etc. your vehicles.

I'm totally with you there.

We're just disagreeing when that point is reached.

I felt the tweet thread was dismissive as well. Hey is running pretty standard Rails things IIRC and I wouldn't consider them niche. They like to make money and they know they can make more of it if they run on baremetal. I think more companies are going to move back to a more hybrid model of cloud + baremetal in the next 10 years. Especially the ones that need more power, flexibility, and cost savings from their infra.

You nailed it on the head. The problems have just moved from one camp to another. We make our life easier because we can talk to an API to get our network, compute, and storage but the complexity in doing so has a pretty high price as well. The current devops trends are mindbogglingly complex compared to most of software that is getting deployed with them.

> On balance, it seems we just swapped one set of problems for another set of problems, and we haven't really gained all that much in the end.

That is the crux right there. And some people need one set of problems fix (so do the cloud) and some need a different set (so do on prem).

I very much prefer on-prem, but for some sets of problems, the cloud does make sense.

Reply on news.ycombinator.com