Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

I got pwned by my cloud costs

troyhunt.com

Post reply on HN
Print view View original

311–320 of 657 posts

Re: I got pwned by my cloud costs

#311

Post by coder543 »

zzt123 wrote: post #18

Interestingly, Troy says that egress is expensive on Azure at $0.014 AUD/gB (~$0.010 USD/gB), but that is the same price as additional egress for Linode and DO, and Linode egress has never struck me as expensive. In fact, I’m kind of shocked (as an AWS user) that Azure egress is the same price as Linode. Actually, wow it seems AWS is also the same price as Linode and DO for egress. While Linodes and DO do come with d…

You’ve interpreted the numbers wrong. Yes, Linode, DigitalOcean, and most of this class of providers charge $0.01/GB. Almost literally an order of magnitude less than Azure or AWS. The megaclouds massively overcharge for bandwidth. It’s not even close.

AWS charges $0.09/GB, and Azure charges $0.0875/GB.

Maybe Troy Hunt gets a discount for being a Microsoft Regional Director and MVP. (Neither of which make him an employee of Microsoft, confusingly enough.)

https://docs.digitalocean.com/products/billing/bandwidth/

https://www.linode.com/docs/guides/network-transfer/

https://aws.amazon.com/ec2/pricing/on-demand/

https://azure.microsoft.com/en-us/pricing/details/bandwidth/

Re: I got pwned by my cloud costs

#312

Post by fbrncci »

rhn_mk1 wrote: post #294
fbrncci wrote: post #290

Earlier quoted context omitted.

I have a pretty complicated architecture that would cost me about 20-35$ if it was hosted just on Digitalocean or Hetzner. Instead its AWS ...soon to be multicloud, and costs me about 140$/mo (which does vary). But it does allow me to experiment, write long articles and design some fun stuff; about which I blog on my own website. The blog has gotten me both clients on freelance projects and enough "cred" to start on…

What stops you from having a blog on Hetzner? That doesn't seem like it has anything to do with AWS whatsoever... or do they offer a blogging pltform?

I'm blogging about the experiments I am running with AWS hosted infrastructure. It could be hosted pretty much anywhere, a rpi would be enough. But I can't run those experiments on Hetzner, they simply don't offer as much options as AWS to run experiments.

Re: I got pwned by my cloud costs

#313

Post by pcthrowaway »

marcan_42 wrote: post #283
pcthrowaway wrote: post #281

Earlier quoted context omitted.

And this is a cloud service is it not?

A dedicated server isn't usually considered a "cloud" service. It's a physical server allocated to you, with unmetered bandwidth and local disk.

Where do you get bare metal servers for that cheap? I assumed by VPS you were talking about a VM

Re: I got pwned by my cloud costs

#315

Post by Symbiote »

hardwaresofton wrote: post #180
dx034 wrote: post #154

Earlier quoted context omitted.

How's that? Setting up a revers nginx proxy with cache takes probably less than an hour even if you've never done it (speaking from experience). And otherwise, if the files don't change that much just ssh in, copy them on the server and serve them via nginx and cloudflare tunnel? I'm in no way a sysadmin and have set up these configurations manually in less than an hour for side projects. Cloudflare tunnel also allow…

Ah sorry I should have been clearer on this -- "global cache" === CDN. Hetzner does incur a performance latency (unless you use the brand new US DC of course, and your customer happens to be in the US). IIRC right now you can't mix US cloud servers and German ones in the same Load balancer (also a relatively new hetzner cloud feature) but of course you can do some DNS tricks and get the loads to be fast. Actually hos…

If I understand this particular case correctly, the large files are just big data downloads of several GB each.

Latency isn't particularly relevant for this, and it probably isn't relevant for most hobby projects.

Re: I got pwned by my cloud costs

#316

Post by bluedino »

Reminds me of a time, we had a new site that was going to run on GCP, we had been using a couple co-located servers for years.

When everything was moved to production, URL went live, nobody ever did any kind of bandwidth checking, caching, no CDN, no cost tracking. $10,000 in our first week. That's about 1/4 what our total spend on the co-located servers was for the whole year. Boss flipped his lid and wanted to kill the new guy who was on the project.

After about 2 years we got rid of all the co-located stuff and were spending about 1.5x, but we had more apps, they served heavier pages, etc.

Re: I got pwned by my cloud costs

#317

Post by hkh »

Olreich wrote: post #304
hkh wrote: post #287

We've been thinking about this for a while, and if there is any way we can catch these types of cost spikes before they happen. We've managed to do it for Terraform resources using an estimation approach, and using a usage file, you can model expected usage-based resources ( https://github.com/infracost/infracost/blob/master/infracost... ), but this one has got us thinking more about policies. To be clear - we would…

I think this is fundamental to on-demand services. Anything outside terraform or another configuration file system is hard to reason about. If cloudflare is in your config system, then you could put up a warning that files bigger than whatever won’t get cached, but that still assumes a level of knowledge about the system that you don’t generally have. Setting up limits and alerts as part of the system creation is usu…

I like that, maybe we have to build up a knowledge base of wisdom (probably learnt through the hard way), and warn if the conditions are met or at least a list of the things to note. Then the cloud cost alert being a fallback safety net.

Re: I got pwned by my cloud costs

#318

Post by Spooky23 »

southerntofu wrote: post #238
lpcvoid wrote: post #56

Can somebody explain to me why I wouldn't just rent a 40 EUR dedicated server from Hetzner with unlimited traffic and gigabit uplink? His 600GB/day is way less than what you get over a gigabit link within a day. Sure, sudden bursts would perhaps "throttle" at a gigabit, but according to his article that was only the cloudflare proxy anyhow, so no pain in having that take a few seconds longer. As far as I am concerned…

Just did the calc and 600GB/day is about 55Mbit/s. That's really not a lot and if there's not too much computation server-side you could serve this from a raspberry pi at home (provided you have good uplink). But that's assuming you keep the CloudFlare cache of course, or as author mentioned himself, advertising only torrents for the multi-gig files. I really don't understand the cloud craze. Everything is more compl…

When you are growing, it’s a no brainer. When you are at steady state it depends.

As a case in point, I worked in standing up a critical system in a large enterprise a few years ago. We spent about $12M on compute, storage, networking, etc. At operational state, it was about 40% cheaper than AWS. The problem is, it all sat there for 6-18 months filling up before we fully hit that state.

With a cloud provider, you pay a high unit cost but if you engineer intelligently your costs should move with utilization. Except for government, most entities generally want to see opex move with revenue and prefer to minimize capex where possible.

Re: I got pwned by my cloud costs

#320

Post by dijit »

bluedino wrote: post #316

Reminds me of a time, we had a new site that was going to run on GCP, we had been using a couple co-located servers for years. When everything was moved to production, URL went live, nobody ever did any kind of bandwidth checking, caching, no CDN, no cost tracking. $10,000 in our first week. That's about 1/4 what our total spend on the co-located servers was for the whole year. Boss flipped his lid and wanted to kill…

1.5x is pretty good.

We overspent quite heavily on our on-prem stuff for a game I helped launch, for political reasons the next game ended up running on the cloud.

The price was roughly 10x before discounts. With our heavy discounts and a wide amount of slimming down/cost optimisation (easily 3 months of work) we got it to 2.3x

There will always be a need for sysadmins/cloudops/devops for that environment, so we didn't save any headcount either.

I can't imagine getting anywhere close to parity in costs, Functions-as-a-service ended up costing more than compute instances too so we went back to compute instances in places where we thought we'd get away from it.

That said, it was a lot nicer to use!

Post reply on HN