Docker operations slowing down on AWS

Agreed. After a lot of struggles trying to find a good average IOPS on AWS for our database, we just increased disk to 1tb with gp2 and got rid of the problem. It gets 3000 iops/sec all the time and you never have these problems. It's still much cheaper than io1.

It's a trade you must account for when choosing cloud providers, nothing is free. For now it's still cheaper to pay for servers than spend time setting things up ourselves, but we're always measuring cost x benefits and thinking if we need to leave aws. The time is coming, but for now it's still working for us.

EBS optimized VMs looks nice on paper, you can choose the size and pay for your needs only. But the once you start to use the disk in production you see the problems.

In short, if you want to use the disk a lot, you need to pay a lot. If your app is slow on aws and uses EBS as storage, increase the disk size, not the VM instance type. This is true mostly for database performance, which once the RAM is filled, relies on IO a lot to get new pages from disk.

I do contract devops for both physical servers (that includes me occasionally travelling to a data center), managed servers in a full service colo, and AWS, and the overall cost per server consistently ends up higher including all devops time for the AWS instance in my experience.

I love it when customers pick AWS (though I usually advice not to, unless they have very specific needs), as my billable hours are way higher for those clients, though it is annoying having to deal with the inevitable "why is my AWS bill so big?" after I'd told them exactly why it'd be expensive in the first place. This is particularly true with bandwidth heavy setups, where AWS charges tens of times more per TB transferred than e.g. Hetzner.

Often I even end up getting paid to help them get off AWS again down the line when they realise how expensive it is.

Overall the idea that managed servers and even bare metal colocated servers take up so much more operations time is just not what I experience. Even if it did, if you're even moderate successful it'd need to save you a crazy amount of operational time per server to make up for the price differences in the hosting.

Further my experience is that you don't need to over-provision much exactly because setting up hybrid setups where you spin up AWS instances (or any other cloud) to handle spikes works well. The trick is to treat managed servers exactly as cloud instances, apart from at most the initial hardware provisioning (though many hosting companies provides APIs for this so you can abstract away that too), so that it doesn't matter where you provision.

As for services like RDS etc., they're a very mixed bag. When they work for you, great, though they're expensive, but very often I end up with clients having to move off them because they need some plugin or other that isn't available. Very few of my clients in the end stay on them for very long. Their biggest benefit is to defer the initial setup of a self-managed cluster.

That doesn't mean there are no cases where AWS is the right choice - for starters having it there for traffic spikes is great (though we usually end up needing it very rarely), and if you need large batch jobs or other environments you spin up/down frequently, it may be cost effective. But it's extremely rare I see cases where it's cost effective for base load.

The typical argument then is that big companies wouldn't have picked them if it wasn't. But big companies don't pay the prices mere mortals pay. I know concrete examples of large negotiated discounts for a couple of larger companies, and they are steep.

> and I'd have to spend more money and time on auditors

That might be true, but most companies are not in a position where they need their IT infrastructure audited. This might very well be a niche that makes it worth using AWS.

> And I've haven't even started to think about securing the resulting systems to the same level I get for minimal effort from a major public cloud.

My experience is that getting security right on the public clouds is harder than bare metal. If you take the effort to do it properly, the end results can be good. But a lot of that is simplified a lot in a colo'ed environment by simply physically separating resources into different network segments and the like.

For people with very complex requirements, you might even get better results, but I could never agree with "minimal effort" - it's very common to see people badly misconfiguring their IAM setup for example, because it was too hard for them to figure out how to open up just the specific things they needed to open.

I think the "5x" (though my experience is closer to 2x-3x) tends to come in if you already do the good/modular design and are set up to scale into hybrid setups as needed to handle traffic spikes. In those instances you don't overprovision, or overprovision maybe 20%, knowing that extending your virtualized setup into a part on premises or manage hosting setup, part public cloud is fast and seamless.

E.g. I had a setup that spanned on-demand instances, rented managed servers, racks in two separate colos and racks on premises. We expanded resources whenever it was cost effective at the time. Generally the colos won out, with on-demand instances handling traffic spikes, and managed servers primarily used for locations we did not have staff.

When your infrastructure is designed so that adding a new one of any of those is just a matter of assigning IP space to the new satellite network and deploy the first instances - whatever they're physically on - your utilisation of all the resources can be far higher.

E.g. in this setup we have instances where we move containers seamlessly between the UK, New Zealand and Germany currently depending on load, available resources, and which instances need low latency (Germany vs. UK makes a roughly 8ms latency difference despite going over an encrypted VPN connection, so we've even had times where client traffic hits load-balancers in the UK while the web servers were temporarily in Germany because it happened to be cheaper to expand there for a while (and contrary to with AWS, our bandwidth costs in both locations are trivial).

If you're comparing to "lets throw a bunch of servers somewhere", then, yes, AWS probably won't be that much more expensive, and presumably that is a big part of why so many people gets caught out by AWS costs once they start scaling up.

If you're interested in IO performance, maybe don't run Docker - whose main advantage over VMs is fast IO - on top of VMs unnecessarily?

Triton and OpenShift add proper isolation to Docker and hence provide fast IO since you're not adding a layer of Xen.