Hacker News Reader

Comment by doctorpangloss | Hacker News Reader

Do you need a tool to tell you that the cloud is more expensive?

Do you need a tool to tell you that your company underpays or is bad at recruiting? Or maybe that you do something boring that people don't want to work for?

IMO Keda is the more important product in this space, because it translates business requirements like max queue wait time into compute resources.

If you are operating in a way where small cost differences decide if you are break even if you have already failed, no amount of "FinOps" will stop your trend from going to zero. It is delaying the inevitable.

View on HN

I've built many projects where "the cloud" is less expensive than the alternatives

The last big one: replacing a backbone country-wide network in France with AWS

It is far less expensive due to the low bandwidth requirements (and also the ability to go full managed services)

What stops AWS from "increasing the price" until it is very close to what you were paying before, if not more?

Nothing, and they do that regurarly

What stops dark fibers or hardware providers from "increasing the price" ? Nothing, and they do that too

So you're proving my point. The tool cannot tell you the single most important factor in pricing.

The single most important factor in pricing is accurately predicting the future, as we haven't invented time travel yet we're gonna call that a wash. Here's a tool that measures everything else.

Yep! You are agreeing with me. The tool measures a bunch of noise, essentially, since it cannot predict Amazon’s pricing roadmap. Whereas the Kubernetes ecosystem has plenty of valid forecasting tools, such as one which forecasts compute usage, that are valuable.

To that point, I believe it's possible to buy spend futures now in cloud costs and if you're doing that then technically it is possible to predict Amazon's pricing roadmap. I know for sure if Amazon themselves didn't (they do) third parties would (they do, DoIT is an example of one I know that does this but there are quite a few of these compute resellers out there).

But sometimes people are not near enough the spend or product support level where it makes sense to do that. I would gander, though, if you're running cloud cost tooling you probably are one of those people.

Competition?

I've never had AWS "increase the price." Is there some reason to suspect this is a thing to actually worry about?

Price is supposed to cut in half every 3-4 years, if it isn't it is largely because cloud vendors can't bear to take less and hope there really is that much more volume of half as valuable computing to do.

The things they sell as an average CPU performance, etc are basically the average when their cloud began.

Moore’s law came to an end about a decade ago. Since the clouds began a vcpu then is largely the same as a vcpu today.

No, Moore's law is fine. If the CPU takes half the real estate and uses half the power but costs the same it is the same perversion of the market expectations of Moore's law as if you don't deliver twice as much in the same price.

From Wikipedia:

Some forecasters, including Gordon Moore,[122] predict that Moore's law will end by around 2025.[123][120][124] Although Moore's Law will reach a physical limit, some forecasters are optimistic about the continuation of technological progress in a variety of other areas, including new chip architectures, quantum computing, and AI and machine learning.[125][126] Nvidia CEO Jensen Huang declared Moore's law dead in 2022;[2] several days later, Intel CEO Pat Gelsinger countered with the opposite claim.[3]

I think it's just because if you believe in economics it would suggest AWS must've lowered the price per vCPU since CPUs have become significantly more efficient in the last decade. Obviously that didn't happen...

> if it isn't it is largely because cloud vendors can't bear to take less

Cloud vendors have to buy products themselves. If their prices aren't changing then we have a market problem that extends beyond AWS.

This is probably why Graviton exists and provides cheaper VMs and Lambdas running on it.

> as an average CPU performance

CPU performance, in the cloud, is effectively free. You can overcommit and time share the CPU.

What you're paying for is memory.

> If their prices aren't changing then we have a market problem that extends beyond AWS.

How do you get there? Even in markets without any specific lock in, the top brands charge prices that have nothing to do with raw costs.

> with raw costs.

You've mistaken my point about "total cost of customer" for "raw cost of materials."

The "top brands" probably do marketing. That costs. They probably have better supply chain availability. That costs. They probably have on site engineering support they can offer you. That all costs.

What you're highlighting is the cost of CPUs has little to do with raw materials and with all this other attendant process. That has been true since a few short years after the product started existing.

> Price is supposed to cut in half every 3-4 years

Citation needed

> if it isn't it is largely because cloud vendors can't bear to take less

Even if we choose to accept your original premise, I think anti-competitive behaviour is much more likely than "can't bear to take less".

New public IP charges and Kubernetes base pricing both doubled my bill.

Spot pricing is kind of a black box, and it also makes no sense.

When did EKS costs change?

They must not have upgraded their control plane :)

You moved the goalpost from "cloud is definitely more expensive" to "well it isn't but maybe it will be for you someday" and alluding to vendor lock-in.

Vendor lock-in is a concern with cloud platforms but doesn't relate to the app in the submission.

It is always more expensive. I haven’t moved any goalposts. What I am trying to show is that you don’t need a tool - indeed you don’t need to look at a single price of anything - to know why. It can be both more expensive and a better value though! If you can’t recruit people or what you do is boring or if you want to operate a business that loses money in the long run: those are some common reasons it can look cheaper when it’s not.

> It can be both more expensive and a better value though!

No, it can't be. Stop with the pseudo intellectual linkedin lingo please.

RDS is an example of something that is more expensive but can be better value - because a lot of the managed service stuff saves the time of those looking after it.

15 years ago I worked an infrastructure department with 50+ employees - these days a lot of that work that we used to do back then is taken care of by AWS.

Scaling the infra to meet peak capacity (which might be two days in a year) because you can only run on hardware you have, having data centres and data centre engineers, are all costs that go away with cloud, even though you are paying more for the compute you do use.

Haa, yes, the scary "vendor lock-in" !

I challenge you : give me a single product that cannot be moved away, given some (cost or time).

At the same time, I again challenge you : give me a single product that you can move between providers with no cost nor time.

To that last one, you can actually find stuff. For instance, you could setup your nginx in instances, without any kind of interactions with the hosting provider. In which case, you limit yourself with what everybody can do, leaving all opportunities to only use the base minimal.

But wait .. are you not locked-in with nginx (or to use a better example: with redis) ?

Vendor lock-in is a scam to increase your bills. Because, in itself, everything is vendor locking. And it does not matter. Think your architecture, design your code, and act when necessary.

I don't think you've ever been through the pain of moving a big, fully integrated, multiple ETL pipeline data warehouse built on something like Redshift. Vendor lock in is more about inertia and less about there not being an alternative. Some things are just really, really difficult to move and ridiculously impactful to change.

You take the term "lock-in" far too seriously.

Yes, one can migrate their entire CI/CD ecosystem, countless IAM policies across multiple AWS accounts in an organization, their lambdas, databases and S3 buckets over to Azure given enough time and money.

The thing is that most everything in the cloud is billed on subscription, and many core components of networked services need not be (except support). Open file server, self-hosted IAM and standards-based execution stacks. Don't like hosting your nginx on Vendor-A? Move it to Vendor-B.

Yes, the tradeoffs are costs. No, vendor "lock-in" is not the only component one should measure when deciding an architecture. Yes, "lock-in" can occur with on-prem software, and hardware, as well.

Yes, it's real and worthy of concern when a vendor knows they can squeeze you because you have no other choice and the level of effort to migrate is too high.

You seem to be trying to reduce the vendor-lockin issue to a binary "locked" or "not locked," but Idon't think that's a productive or realistic way to think about it. It's really more of a spectrum, and I prefer to think about it as "options" rather than "are we locked." You definitely have a lot more options if you go with Nginx and the project turns evil than you would if you went with some proprietary offering.

> Do you need a tool to tell you that the cloud is more expensive?

Yes. If you want to make an argument for change in a business environment you need numbers.

It’s the wrong question usually. The right question is how many more features and what revenue growth would you get by using cloud? How many less people would you need to employ? Would you need to manage facilities - worry about hardware, power, diesel generators, etc.?

Cloud is more expensive. But you’re not stuck with capacity and sysadmin tickets.

> Do you need a tool to tell you that the cloud is more expensive?

More expensive than what? Doing your own?

That depends on what you are doing. Often, the cloud is cheaper if you factor in the human cost. I've seen many spreadsheets that just quote the hardware costs.

Not only is it potentially cheaper, it can be better than rolling your own stuff in-house.

Kadokawa can tell you all about how their in-house shit (literally, Japanese suck at computers) got pwned while the stuff they had in the cloud (I hate the term, but it is what it is) were unscathed.

Keeping in-house has its benefits, but the costs are also steep.

This tool makes it relatively easy to find cost savings, and also to do chargebacks. Most kubernetes clusters waste tons of compute for no reason at all.

One of the main value adds of these tools is to help identify and categorize workloads into OpEx and CapEx buckets

Unless you are currently in possession of a bunch of latest generation NVIDIA DGX, which many people are, 99% of your workloads are OpEx.

They might be CapEx in some imaginary, nonactionable accounting sense. Like maybe you are mislabeling exclusive rights to dig holes in some neighborhoods as telecommunications equipment capital expenditures. Just pay an accountant to make up stories for you. You don't need the tool.

Reply on news.ycombinator.com