> As a startup you are typically better served by a DigitalOcean-level of complexity, and there are plenty of such offers in the EU (Hetzner Cloud, Gridscale, OVH, etc)

As an actual startup founder who started as a 1 man startup, strongly disagree.

Spent maybe $200 a month on Google Cloud, got an actual production ready cluster. Scaled up to Millions in revenue, never had to deal with any Linux Server admin BS.

More time on business, less time on Linux Sysadmin.

> never had to deal with any Linux Server admin BS.

Oh, you just had to deal with a different flavor of BS. Or you was lucky and everything just worked out for you (but why Google Cloud and not some PaaS like Heroku, so you don't have to deal with cloud infrastructure/servers BS altogether?)

I've been both a system administrator, managing GNU/Linux and FreeBSD servers in the ancient ages, and DevOps guy doing all sort of stuff in the clouds. The complexity is still there, it hadn't disappeared in some magic cloud pixie dust, even though sales would wanna tell you that fairy tale. But here's the thing - you never get to dive into those waters (or hire someone to do it for you, be it an employee, contractor or paid support) unless shit hits the fan and forces you to.

You must've cheerfully walked through a minefield and haven't stepped on and even seen any mines. Honestly, I'm happy it worked that way. And hopefully, this minefield is sparse enough those days so you're a rule not an exception - I don't have meaningful statistics. It would be actually interesting to run a poll or something. I just happen to have seen a few companies/people for whom clouds weren't all unicorns and rainbows.

And as for the flavors - it just happened that you knew how to set up stuff in Google Cloud. Would you happened to know how to spin a simple instance on Digital Ocean instead and went that way, and be lucky to not encounter any serious issues, it would've been the same painless experience, just different flavor.

My server load was not the size I needed 20 dedicated servers, but far too much for Herkou. Just running a 120 core 24/7/365 on heroku is like.. all of my revenue. (Vs 1% on google cloud and maybe .1% on hertzer).

100%. The hidden part here is "DigitalOcean-level of complexity" is actually "DigitalOcean-level of features."

The big cloud providers have a variety of offerings of different complexity. Using GCP as an example: want k8s with all it's flexibility and complexity? You have GKE. Want to still run containers, but abstract away all the cluster resource management? CloudRun. Abstract away the container itself? CloudFunctions. AWS has EKS, ElasticBeanstalk, etc.

I understand people get overwhelmed the first time they're dropped into the console of these cloud providers but really it just takes a bit of reading to figure out what you should/shouldn't care about. And the benefit of doing so is enormous.

Disclaimer: Anecdata

Privately I host nearly everything on a shared host in Germany (that is everything I can host without sudo) [1].

For company policy reasons I must absolutely use AWS or GCE.

For an internal project I need to setup Matomo. Something I did thrice in the last few month on [1].

OK login through SSO into AWS. Look around, ask Google, find the bitnami image, click few buttons. Done. OH shit. Now I need to somehow make it publicly available. OK. Google again. Ah this is the way. Few hours of reading and clicking later I have a publicly reachable Matomo instance. Oh hey. It warms me that it is not ssl encrypted. OK. How to do let's encrypt? Google again with my second batch of coffee (or was it the third). Found an easy way, just enter a command in the shell. Oh hey, how do I get my ssh pub key into my EC2 instance?

Damn the day is nearly gone and I have yet to deliver this tangential asset to an internal project while killing my CCI (how much I am booked on client work) for something that the first time took me 30 minutes with the great documentation from [1].

To me as a meager Data Analyst the complexity of cloud offerings is a nightmare. And the documentation is written for other echelons of tech understanding most of the time.

[1] uberspace.de

If you’re a data analyst, then of course infra and sysops activities on cloud seem complicated. I’m sure a sysadmin could run/write sql, but would find the rest of your domain complicated too.

OVH gave us home. Enrolled us to Startup Program. Gave us support and hefty credit.

Managed K8s. Openstack.

When we started paying for it, it was still cheaper than AWS.

Just because AWS is the default, does not mean you should use it.

Was that before their datacenter caught fire and their customer servers were lost or after?

Yes, OVH experienced force majeure episode. I didn't follow exactly how the compensation was rolled out. I know it was messy. I am not going to defend their actions, I am sure they could always handle this better.

Disaster recovery planning is practice we should all adhere to. Hindsight is 20/20. Not trying to be a smartass. I know it was painful for a lot of folks.

At the same time, unless you paid for managed service with clear SLAs, then responsibility is yours.

Cloud is just someone else's computer.

FYI: we started with OVH before the fire

When I started, no other providers had the K8s features I needed.

Still prefer Google, as they are the OG for k8s.

And power to you. You did what you though was best in your circumstances.

Today circumstances have changed. You need hassle free scalable DB, then AWS RDS might you best choice. Maybe.

You need open standard IaaS, well, there is ton of options.

Even before K8S, you had and option of Openstack with Ansible. Yes, very different beast, but still much _simpler_ and _cheaper_ than stocking on large number of IT professionals.

We colocate about 20 servers and in any given month, spend no more than 1 person-days worth of time dealing with it. Many months we spend no time. That includes both sysadmin and hardware. But this requires knowledge that most devs these days probably don't have.

We might spend more time messing around with AWS than our colocated servers.

Right, if you legit need 20 servers than it might make sense for you. I would fit on like 2-3 decent sized servers if I did co-location, and would save not even 1 developer-day of salary...

You can rent 2 to 3 dedicated servers at affordable rates.

Right, but my time is worth more than what I could possibly save at my scale. You have to find the right balance.

I would guess that if costs is an issue then it must also be balanced compared to the potential profits. If your current $200 solution only allowed you to have US customers, while a $300 solution would allow you to have both US and EU customers, which one would you choose?

>I would guess that if costs is an issue then it must also be balanced compared to the potential profits. If your current $200 solution only allowed you to have US customers, while a $300 solution would allow you to have both US and EU customers, which one would you choose?

Whichever one let me pay rent at the end of the month

The US has plenty of customers. If I had to drop EU, I will. It's a nice bonus, but not a core requirement.

That seems good. Someone could copy your business and spin up on the EU market. If its profitable its profitable and its no worry for you. If its not profitable then the EU market is not large enough to carry the product on its own. GDP of the US is around $25 trillion, and EU is around $18 trillion, and population wise there are around 300 million people in the US and 400 million people in the EU.

Might I ask you what kind of product your 1 man startup have?

I am not 1 man anymore, we grew up a bit. But we are an ecommerce platform, basically centered around the big US marketplaces (Amazon, eBay, walmart.com). Yes Amazon and eBay are in Europe, so we are there.. but no say UK or France specific markets at this time.

Just make sure your BCP plan includes other provides. HN is full of stories where peoples' accounts are blocked with no reason and without means of effective contact.

That is consumer accounts, not business accounts at GCP. People confuse personal gmail with paid GCP. I have actual reps I can talk to.

Tangential, but assuming you're talking about Listing Mirror? I considered working on a similar product a few years ago, but felt the market was too competitive. Interesting you were able to compete with the plethora of similar services.

Indeed.

Pros and cons to being in a crowded market.

From day 1, you KNOW there is demand for your product. You can look up Channel Advisor and see the revenue. And 20 smaller companies under fighting for the rest.

Cons of course being, you have to figue out how to compete with all of these guys ;)

Working on a consumer SaaS startup and I strongly disagree. A virtual machine on something like DigitalOcean does not provide any of the nice abstractions that something like a Google Cloud Run (or similar on AWS/Azure) provides. The amount of time a cloud provider can save you administratively is difficult to exaggerate. That is from day 1. Should your startup succeed, and you need to scale, the real savings start to kick in since scaling to a large degree is handled for you. Good luck re-architecting your app into a kubernetes cluster and handling load balancing manually while your competition gets all that with almost no effort.

I've worked on so many small teams where dealing with AWS/Azure etc was a huge part of their day, for very very simple products.

I still remember arguing about bloating a web app with a 1mb package from AWS so it could use their serverless authentication offering.

Common theme as using those lambda function - sometimes paying quite a lot of them - to serve requests that would be twice as fast on the proverbial $5 linux instance.

So yeah, looking from the sidelines it feels like a huge amount of added complexity for small teams, "just in case" they need to scale. Which given how fast modern hardware is way further off than they think.

(unless they use lambda functions for every API request. in which case they better learn to scale in a hurry)

All Office docs contain a GUID that's closer to PII than an IP address.

Can't find any references to that except that each doc has a GUID. A GUID on its own is not PII - just some random number, so are you implying that MS collects every GUID along with author identifiable information?

I just got off the phone with a lawyer to talk about this exact issue.

If the GUID is related to the user (like user ID), then it is Personal Information - EVEN if the GUID is random. The distinction that is easy to miss is that a User ID GUID might be very low risk (compared to, say actual User Id or user name) - but is is still Personal Information.

If the GUID is for the document (and anyone can edit the document), then it is no longer PI.

Of course, all of this ignores things like the contents of the doc. If the doc is "SSNs of my customers", well... don't do that

The free credits are nice, but run out quickly. Azure is expensive, but has a lot of nice tools from key vaults, log analyzers, CDN, databases, pipelines, to firewalls. I mean, yeah, you can implement similar stuff on a DO platform but you're going to be wiring it all up yourself, taking on the liability for keeping it all secure, and providing the warranty for its availability and effectiveness. The value to AWS/GCP/Azure is far beyond free credits. They've commoditized services - it rarely makes sense to pay for in-house expertise in managing those services yourself.

Also, the offline version of Office is going away, to my knowledge. I think the current boxed version is the last boxed version they plan to sell.

I would say that looks at the state of things as they are today, which may not be the case as technology advances. If there's a service that provides a real competitive advantage that is only available outside of Europe, then this might exclude businesses in Europe to innovate and compete.

Stockholm, Sweden begs to differ as they just dropped Office 365 due to Schrems II: https://www.version2.dk/artikel/microsoft-forsikrer-lovlige-... (Danish article)