Hacker News Reader

Comment by ajross | Hacker News Reader

What's interesting isn't the fact that they "make" more from AWS, it's that AWS gets premium margins where their store is running right at cost (and has been, basically forever -- it's their thing, sorta).

The interesting number you left out was the non-AWS revenue, which in 2018 (too lazy to correlate with today's announcement) was $230B, literally 30x higher. AWS is a tiny drop in the bucket, no matter what the profit details say.

Which then prompts the question of why AWS gets that kind of margin, when it would seem like a pretty commoditized field at this point. Linux VMs are an open and very portable deployment platform. You can get them from any number of top tier providers and a huge horde of little ones. You'd expect to see a ton of competition driving prices down, but we don't. I genuinely don't know why.

View on HN

AWS is basically "run your entire company's engineering as a service" now, and they've done a good job at it. It's not at all easy to compete with such an integrated and battle-tested platform (other than on cost; AWS can be pretty expensive).

This is absolutely true, sure, but the UX is dreadful. If someone like, say, DigitalOcean who has nice design spent the money to actually compete, I'd move in a heartbeat. Google is a nonstarter, and Azure is a 'maybe'. Most people use a few key services. Everytime I log in, I'm slammed with 200 tiny hyperlinks.

Where I work we spend millions per month on AWS, its our primary cloud platform.

At that scale the UI of the console really doesn't matter, everything is IAC so developers rarely need to directly access AWS. The one thing we do use is the Cost and Usage explorer which has a reasonable UI

> UI of the console really doesn't matter, everything is IAC so developers rarely need to directly access AWS

Okay, but those "rare" times when you do need to access the infrastructure, having good UI is pretty helpful.

Just curious, why is Google a nonstarter?

Most companies don't even consider Google, it's a distant third and provides very limited offerings. There is a lot of love for Google on HN, but reality does look different.

Ther is also a common believe that they will just drop support for something whenever they feel like it - which in regards to GCP could be unwarranted but well it is what it is. I would pick AWS first, Azure is becoming a close second and GCPa distant third. I think Digital Ocean looks awesome and might be worth a look before GCP.

Also, GCP is having stability issues recently.

Fear of abandonment perhaps.

We are looking to move some of our applications to the cloud and Kubernetes the coming time. We'd like to go for DO, partially for the horror the AWS interface appears to be. Could you give me an idea what we'd be missing by choosing DO over AWS? (as your comment appears to say we would)

Just of the top of my head you miss the following - whether they matter to you is a question only you can answer:

* Access control for everything via IAM. This for me is the killer feature for AWS, almost anything from users, to servers, to individual IoT devices can be granted permission to access other AWS resources with extreme granularity.

* Audit logging of every single API call, user invoked or otherwise, via Cloudtrail.

* Monitoring of and responding to key metrics via Cloudwatch. This isn’t just graphs as it can appear on the surface, Cloudwatch alarms allow you to do things like killing individual servers if they start throwing errors while others aren’t.

* Responding to changes in your infrastructure via Cloudtrail. We’re using this for features ranging from emailing our security team when a new IAM user is provisioned, through to full release orchestration.

* All those features supported across services that can fulfil more or less any requirement you have, whether that’s a MySQL database, a message queue, or (should you really need it) a satellite downlink.

DO are fine if what you want really is just a server in a data centre somewhere, but you’re missing out on a lot of really powerful management features with them, without even getting into the other services AWS have.

DO provides VMs and Kubernetes as a service. AWS provides Freaking everything as a service. You might need Redis. You might need Postgres. You might need Hadoop, NFS, A Monitoring Solution, Access Controls for your account, Docker Repo Hosting... AWS does all of these things as managed services (though some not very well).

You definitely don't want to run your RDBMS on Kubernetes, and while DO's "Marketplace" has a solution, it's not the same.

DO has managed hosted Postgres now, with Redis and MySQL coming. I’ve been using it for a few months and it’s been fantastic. The interface to get it running is unbelievably simple. DO is killing it lately.

> You definitely don't want to run your RDBMS on Kubernetes

If you’re targeting AWS you’re probably better off using their solution, but you can totally run your database on k8s. And for some uses it’s even a good idea.

If you're new to Kubernetes, you shouldn't attempt to run a RDBMS, especially if it's not a dev/staging DB and/or needs to run in a HA config. Operators aren't mature and if you're rolling your own statefulset you better know what you're doing (with kubernetes and with the DB you're running).

> If you're new to Kubernetes, you shouldn't attempt to run a RDBMS ... if you're rolling your own statefulset you better know what you're doing

I've been doing this for a while. Before there were statefulsets, it was pretty painful. But now, it's relatively easy. Sure, not as easy as a PaaS database that seems to just work, but much easier than managing a cluster of servers with all sorts of crazy scripts.

> We are looking to move some of our applications to the cloud and Kubernetes the coming time.

Google Cloud has a ton of built-in support and UI for Kubernetes clusters. They make it really nice. It's roughly the same price, so I'm not sure I'd run Kubernetes on any other service.

Have you compared it with DO's Kubernetes UI and support? Is DO really worse in that area?

> Have you compared [Google Cloud] with DO's Kubernetes UI and support? Is DO really worse in that area?

Nope, wish I had time to devote to such a comparison. Would love to see an article.

All of the Digitalocean SF data centers were down for at least 10-20 minutes earlier this week during my company’s peak hours. They seem to be having growing pains. I wouldn’t move any critical stuff just yet. I’m optimistic for the future though.

Why is Google a nonstarter?

These parent kind of comments are like saying: “apples are cheaper then oranges. I dont like red cars.”

There is nothing to be discussed here. If you compare do with azure, google and aws and are talking about the interface being most important: you clearly are not the enterprisish client aws/azure/google is aiming for.

I’d go further and say that in almost every domain, there is a startup to be built that can be described as “like X, but with a good UI”.

Never underestimate the power of good aesthetics and design.

> Google is a nonstarter

That might be unique to you, I've been using google and I've very much enjoyed the experience, why doesn't it work for you?

Same thing as a fast food burger versus a restaurant porter house, you get what you pay for.

Analogy is a little off.

AWS is fast food and expensive because consistency, availability, and volume are what count in the cloud business. You can get better quality VMs for less money but you won’t be able to provision thousands easily and it’s not going to be available 24/7.

Even further, I'd say AWS is fast food that supports dozens of niche diets - pretty much anywhere in the world, anytime.

McDonald's isn't impressive because they can make a Big Mac in 30 seconds. They're impressive because I can get almost the exact same Big Mac anywhere in the world.

I don’t think it’s commoditized. AWS has moved past being just EC2. Virtualization is a small piece or the picture, and building and operating DynamoDB or S3 or SQS are all very tough tasks.

It's not commoditized at the scale Amazon operates it at.

And from an AWS / profit perspective, running the rest of Amazon is essentially what gives them a base consumer to get them to that scale.

Which is the strange part. Because it is rather commoditized on the level that 95% of their customers are by themselves.

It's that aggregate of so many individual clients that makes it difficult. So why aren't all the small clients getting picked off by the hundreds of small competitors?

My take is that the sheer scale of AWS has hit a point where engineers (the ones making infrastructure decisions) and moving from one company to another are just more likely to be familiar with AWS at this point. The same concept makes hiring easier.

My other guess is that lots of engineers are like me and just got sick of their hosting company getting bought out and constantly having to deal with console changes and migrations.

I am with GCP now, one of the reasons I picked GCP is because they are very large and have reasons to continue offering cloud services even if it becomes commoditized and unprofitable. The same is true for AWS.

Also, of course "no one ever got fired for choosing AWS".

And RDS.

At my scale two DBAs + MySQL on EC2 is cheaper than Aurora. I believe other companies will come to the same conclusion in time.

Where do you find such cheap DBAs? Are they by chance also in other roles?

You’re comparing quarterly revenue from AWS to annual Amazon? The proper ratio is more like 7-8x, not 30x.

It's not commodotized at all. They have successfully achieved the jump to be the "nobody got fired for choosing company X" of cloud computing. Everybody trusts them. I've been directly involved in planning out cloud migration for an organization and put forward other options that were substantially cheaper but it didn't matter: they wanted AWS.

Simple.

Server multiplexing.

Let's say I increase fleet utilization for 1%. My projected margin increase 1%, not retrospectively though. Plus the projected server hardware price drop. The margin increases naturally. If there were no competitor from Google and MSFT. Amzn would need to be extremely aggressive to not making a log of money.

That's why Google was very confident in winning the Cloud war back in early 2012-2014, and prior to that time frame. Google can operate fleet utilization way higher than industry standard, and clearly is ahead of Amzn then.

But it turns out a business is not that simple as writing code. And obviously, Amzn catches up in fleet efficiency. AFAIK, now Google and Amzn are almost head on head in that department, while Azure is the poor kid lagging behind. But that's not a problem for MSFT. Their strength is in enterprise and conventional developer community building.

Reply on news.ycombinator.com