Hacker News Reader

Comment by AllanSavageDev | Hacker News Reader

I spoke with some high level folks at a very profitable private company recently and inquired as to why they had DBAs on staff for what amounts to a pretty simple OLTP type system. Id naively assumed that someone of that scale would be using a cloud provider (RDS for AWS etc) thus negating the need for someone who really knows DB internals and upgrades and OS level stuff.

The answer was that they simply didn't trust GCP or AWS or Azure to see their data and know how much silly money they were making in the niche industry they almost completely monopolize.

I recently interviewed with a lower-case-m megacorp in a similar situation and they host on-prem for the same reason, at great expense and hassle in facilities all over the country.

Seems like theres room in the market for some kind of an On Prem Private Cloud Stack that emulates GCP/AWS etc but locally maybe?

Cloud-In-A-Box anyone?

View on HN

> Seems like theres room in the market for some kind of an On Prem Private Cloud Stack that emulates GCP/AWS etc but locally maybe?

It's more like cloud-in-a-rack, but that's what https://oxide.computer/ is trying to do isn't it?

From our perspective I'm not sure the cloud abstraction is better or we even want it to be done like that locally. Look at S3 as an API for example. It's absolutely dire. I'd rather use NFS (!). And stuff like Lambda and Cloud Functions are just cat turds.

On the DB side, I can't say too much as we're a pretty obviously identifiable AWS customer if I give out any details. I will only say that nothing fits our size and scale so we have to run on bare metal. That just makes it really fucking expensive colocation.

Lambda is awesome .. until you actually try to use it for realsies. Cat turd is an apt description. Just being able to get the damn logs for debugging is itself a hassle. Terraform helps a ton in all this and I rarely find myself using the AWS UI anymore. Still Lambda is a great idea that just doesn't deliver for any use case more than responding to some S3 upload action or Event Bridge operation.

Don't even get me started on the API Gateway sitting in front of a group of related Lambdas. Its OK once you get it setup and running but buildign/changing it amounts to stabbing yourself in the eyes.

I saw you post "lambda is awesome" and was going to reply with "only in certain circumstances." But you beat me to it.

After about a year of poking lambda to see how it actually worked (versus how it was documented) and building a cloud formation replacement (TF eventually did what I wanted, but not before I made a simple PROLOG based replacement for it). But I finally made it work after MUCH struggle.

So... +1.

I put this in our code several years ago: "WARNING: Do not look at Lambda stack with remaining eye."

You are likely describing software like Proxmox which has quietly been able to do so much for a long time.

Open source, and works great when small, and at scale.

https://www.proxmox.com/en/products/proxmox-virtual-environm...

https://www.youtube.com/results?search_query=introduction+to...

Those local options exist, and have been around forever, but the problem is nobody is doing it without cutting corners and with pay-as-you-go elasticity (and the 'call an API, get a VM instantly' effects that go with it).

Most on-prem deployments were trash and a lot of them still are. Not because it couldn't be better but because it's easier to just have some random hypervisor department do this work manually and not do the work to create it as an internal product. Even VMware with vrealize failed and that's about as 'customisable cloud platform in a box' as COTS enterprise software can get.

Maybe it's because IaC and APIs were just not in the vocabulary of the average system integrator or on-prem operating team (it's still lots of clickops and copy-paste).

Sounds like HPE GreenLake.

Reply on news.ycombinator.com