The REAL Cost Of AWS (And How To Avoid It) 

2024 list will have zcloud. Cheapest and easiest together ❤

That 📛 was a expected scene. This took place

I was just passing By very respectful video just being assertive and land people know that you know you stand by your work you know respect I don't even know that much about holding and running SS3s except you know I've been using what they wasabi instead of you know Amazon cuz Amazon is just overexpensive man that AWS is just man it's just a key towards bankruptcy

Keep in mind that AWS is where most of Amazon's revenue comes from. It's literally where the money comes from for them to undercut other online stores.

great nerd out

maybe you should start looking for a better job (- ‿◦)

He's even more devastating than chatGPT 😂

He's saying the quite things out loud. 😅

Jobs reduced because company picked aws? Zero

Paying Engineers after ChatGPT plugins = stupid!

can you expand on how you did that?

​@@123du5, nothing fancy at all. VMWare VMs and Jenkins build server. Could use any other build server, of course, there are many that are free for self-hosting.

Do you do regular backups? Does data loss concern you?

@@fawazsullia5620, no, there is nothing on the build server / remote workstation that needs to be backed up, aside from the configuration.

there are industries where cloud computing makes poor fiscal sense

@@manonamission2000 It doesn't make sense for many reasons, cost being one of them. It only makes sense for a small company that needs enterprise grade resources to do proof of concept or development. Otherwise, the costing is at least double ( and usually upwards of 500% or more) the cost of doing the infrastructure in house. It MIGHT make sense for a company that needs large scale resilience, but uses very few compute resources. Might.

​@@timothyblazer1749even for small businesses there are tools to setup your backend on the go with very little time and resources. In vast majority of cases using the cloud makes no sense to me.

kinda is right word and then they invented SnapStart

Vercel does

It’s literally the business model of all these vendors. Build a UI wrapper for AWS, have a free tier, and take market share from AWS. Turn off free tier and try to get acquired.

💀💀

I feel like this is why Kubernetes is in so much demand ☺️ a well design container setup can save you in all areas of the business.

I don't get it. If your application is completely serverless, how can you have scaling issues or leaving unused resources sitting around? Serverless is entirely about not caring about those things.

@Aden Rius if you leave a dynamo dB table sitting unused with high wcu and rcu... you get charged whether or not u access the table or have data in it. People just starting out may not understand auto-scaling. My comment is about devs just starting out and not fully understanding the cost implications until they get billed.

@@adenrius Simple scaling is utopia. Say you have lambda's processing some events. If you start getting too many events, your lambda's will scale up to the account level limit - 1000 concurrently running lambdas and it will starve other, non-related lambdas. So now you'll need to decided what to do about it. Say you decide to set concurrency limit on your lambda to 5000, then write a support ticket to increase AWS account's limits. Now you think you're good to go and your architecture is "scalable". Wrong! For example, you might reach throughput limits of your SQS queue. Then you'll need to decide what to do about it: enable high throughput SQS queue, migrate to something like Rabbit MQ or maybe use streaming solution, like Kafka. Is concurrency solved now? It doesn't. Now you could bottleneck your data storage. You'll have to go though capacity planning again and work on a solution for your particular application. And even then it won't be solved forever. You'll need to think about rapidly changing concurrency. What if you need 5k concurrent lambdas right now and next moment you need 20k? You'll need to write some automation to change it like that. And even then, do you need to scale infinitely? What if cloud cost would be too much? What are the external factors that scale your application under the hood? Maybe you shouldn't scale into the space and should define some sensible limits. But then what would you do with unprocessed events? Drop them? Use a dead letter queue? Maybe you'll setup some notification system for developers to react to scaling beyond what you expect. And it doesn't stops there. Scaling is ultimate problem, which cannot be "solved". You'll work on it forever.

@@cloudboogie Exactly this. You need to pick an upper limit... ie "We need to execute a maximum of 300000 widgets per hour" You then need to balance all resources to meet that level (ie, a repeatable performance test) so you are not guessing and optimizing your costs.

Don't they all use AWS/GCP/Azure under the hood? But yeah, I'm also impressed by how cheap especially planetscale looks

When infrastructure is cheap enough... You may as well try to capture the future big companies and get them invested in you. It also helps that a lot of these companies have funding and htey can afford to have a more loose attitude.

@@jocke8277 AWS can't implement a good caching solution unless you set it up. Planetscale does look good for serverless solutions.

That's the underlying concern I have when I use free tiers, to happen what happened with Heroku.

​@@picudo it will happen to qll of them that don't crash. Especially with this recession when all this vc money drains

Yep he just totally skipped over planetscale having those hard storage limits which just kills you if you have any substantial historical data.

have you ever seen a promoter speaking about -ve things of product they are promoting, if they did it will be sarcastic.

or use supabase

did he conveniently leave it out? Or did he simply omit it because he knows that people with edge cases like yours are going to find solutions that fit their needs while 95% of the people watching this channel will never need something beyond the free tier?

Most of my hobby projects I deploy on my home server. I anyways have it for NAS - the cost per terabyte (hardware + electricity) is much cheaper than cloud storage, not to speak of ethernet transfer speed. And once I have an always-on home server, I can easily spin up my hobby projects there as docker containers for no extra cost. (I manage the containers with Docker Compose. No need for Kubernetes and other accidental complexity.)

I think the point is that you can spend _a lot_ of $40/100gb before it becomes worth investing in paying someone to bring that cost down. Like yeah there is a point where it no longer makes sense to stick with these platforms. But by then your app will be making big dollars and moving is justified.

@@JoRyGu this

If you want the same level of availability as AWS does per default you usually need to go with the Enterprise plans with the Vercel types of companies which is significantly more expensive. Multiply that with having to use 5+ enterprise versions of various service providers offerings to get features that AWS has built in for a slightly more complex service than a simple web site. Depending on what services you go with you could easily be spending 4+ developer salaries worth per month more on those subscriptions. You could have invested that money into a 3 person ops team that work full time on figuring out how to do stuff on AWS or hire some AWS consultants do just do the figuring out for you.

I mean.. this is kind of the point of these projects. It's like Heroku. It's amazing early on in a project (and I've used it for several hobby and not so hobby projects). But if you use it to run your production infrastructure, at some point scaling it up becomes almost geometrically expensive. Moving to AWS/EKS got us 50% cost savings, gave us a lot of extra tools to ensure security, compliance, and let us build some really cool dev tooling. However, the company ran perfectly fine for 3 years before hiring a single DevOps engineer just by utilizing Heroku, highlighting its value.

​@@JoRyGu It's not about the initial setup, it's about spending your time on the core business issue you're trying to solve. Everything else is waste, Google lean manufacturing, Toyota they came up with this idea. In the end the only thing which will set your application apart from the rest will be that feature or UI that's perceived to be so much better by your main audience, your end-users, that your focus should be there. Because that's where you'll make your money not by saving $20,- a month.

came for this comment

what happend?

@@mauroquinteros8783 planet scale completely removed the free tier

exactly. it is very easy to be baited to use them. They are hard to migrate out off once you grow. You are also paying much higher price than bare bone vm.

I assume you are talking about the server and not the db. Or do you deploy a db on Docker as well?

​@@michaelnurse9089 docker as well, the dbs (usually multiple) are just services, like everything else

yes, but there's buildpacks, too also migrating only needs to be done for 1% of the projects, most of them would not scale until this point and if you do, you have shitton of money, and you can afford devops

@@taylorchu It depends on your need I guess. This is sometimes true but sometimes the opposite. For example on GCP Cloud Run even with the CPU always allocated you would pay 20% less than a regular VM. It can be even more cheap if you can keep the CPU idle when there is no request. It's all based on a Docker image so almost no vendor lock in. If all you want is scram a bunch of stuff on the same tiny VM, well in that case you're right and it's the wise decision to make until you outgrow that setup. I always started there. No need to overengineer.

$240 per developer per year vs. $200k+ compensation per developer per year? If it saves more than one or two hours of work it’s more than worth it.

@@PeterDolan Yeah of course, but Vercel also charges $40 for 100GB-Hour for serverless functions (~1.1E-4 per GB-Sec), that's 6.65 times the price of AWS lambda (not including the $0.2 charge per 1M invocations). If a tiny startup pays 5K a month for AWS, is it worth it to pay 6 times more for a few hours of dev time? Not sure, hard decision...

AWS is actually a better deal long term at any scale

Thats assuming they dont want their business to evolve.

​@@surreal5335not every business evolves fast. sometimes they stay the same for years until upgrade is wanted.

No shit... did you not even watch the video? He talks about the "line of prime" when you get to the point where things are stable and you have a succesful business that it begins to make sense to start isolating parts and rewriting or migrating off to dedicated infra. Most projects anywhere are gonna fail. and for the first few years the important thing is speed and ability to iterate and try new things. This services are all great for incubating ideas at the start up level.

I was gonna say exactly the same. Would love to hear how Theo addressed this problem. Even with such generous free tier, bandwidth will bite you hard in the end

I think what he meant was if your company is already having huge bandwidth to sustain then you should indeed transfer to AWS or Azure. I did something similar to this on a side project where certain functions cost a lot and would be better on lamda to just pay for these really traffic heavy functions while also making use of vercel’s generous free plan for rest of the app.

Have you checked out bandwidth costs for other CDNs? Cloudinary and similar should give you an idea of how reasonable that price is for a global CDN

@@t3dotgg Digital Ocean CDN cost 1$ for 100GB and also 1TB for 5$ included

Clouflare pages is a nice alternative

Same I found this video by chance, I dont realy do hosting ever in my life, I know some word but thats it. Though, I decided to write the names he gave: vercel, etc In case one day I want to make a website with a database (this seems to be the subject of the video)

why ? because of planetscale?

Older? guy, who remembers being excited to get a 1200bps modem! Forget a T3? Most companies 20 years ago had a T1 and that seemed like incredible bandwidth.

Today's guys call themselves software guys but dont want to touch a server. I have setup one rack full of servers on-premise in my home office to do my HPC stuff. Yes its tough but works several magnitudes better compared to cloud. The same capability on AWS will make me lose my shirt in just 2 months.

@@gizmoitus 1200/75 or 300/300 *full duplex* (8N1 hahahahahah) OMG kids these days [weeeeexhxhxhhrhrhhshshhfhshshh Hey there's someone on the line]

@@programmer1356 Love all that geek stuff -- setting up your own terminal software. Poor kids will never know the excitement of connecting to a BBS, downloading shareware, or spending $50 for an hour of compuserve. I had a Netcom account and was thrilled the day I manged to start up ppp so I could run mosaic and open up an actual web pages I had heard rumors about. Not to mention a compiler was $500. I think I spent something like $60 bucks to get a copy of Turbo Pascal, so I could have a "real" compiler to program with. Put me on the road to a pretty good career.

an engineer moving into sales...

This. A lot of these companies are running on top of AWS, that the very basic thing you need would rather be spent learning AWS for a month, if you can afford to. It's not terribly hard to setup similar, although basic facsimiles. My setup is so you can spin up a lambda every time you cli create a folder. And lambdas are cheap, so you should use it for as many things as possible.

Oh man you're gonna love this...

why serverless is nice and easy, just use SAM

Firebase auth / identities worked ok for me and it's free. I'm in the < 1000 users category however and don't expect it to grow much. One thing that is annoying is getting the "email_verified" to true. The verification emails often don't make it to the user's email. I have to call the endpoint myself and click the link.

@@saiv46 You don't choose the circumstances in which you have to work. Moreover, if you do not break the law and do not engage in politics, then nothing threatens you, and this is the vast majority of businesses.

@@titaniumhocker ha-ha-ha seems you are not living in Russia, as you say such things

@@jeremyburton3277 Yes, omniscient USian, please share more of your knowledge about how our country works that you pulled out of your arse

but planetscale does not lock you in it's all standard mysql

​@@titaniumhocker Doesn't seems that much different in the west to be honest. You run a business you don't run politics, people have short term memory and probably forgot what happened with Parler an alt right leaning social media network. If you want to run politics you just need very very deep pockets and 6 zeros in the black on your bank account aint gonna do it.

to anyone curious to play around with Authdog, we're officially launching our beta May 1st 2023 :)

Hosting on the cloud for anything is just a disaster waiting to happen for any start up without massive cash reserves. It is also laughable when this guy said that the edge is also better because it is cheaper, without clarifying the huge drawbacks to using the edge which is why it is being discontinued. So if you had heard this guy a year ago and went on the edge with vercel, you would have to undertake migrating your back end to a regular server, using up any savings in time and money in the meantime.

If you feel robbed of annoying experiences, I'm sure you can find some more to compensate with all the time you just saved

@@noahwinslow3252 aws amplify is ridiculously simple and learning a bit more honestly takes maybe a bit of study. AWS.=, GCP, and Azure already outcompete vercel

Amplify is simple but with a little more work you can deploy the underlying services yourself much cheaper and each deployment will be faster.

Agreed. Learning self hosting is valuable learning experience.

We as a developer has to give up wanting to be proficient at entire tech stack. We simply don't have enough time to learn all that. And job roles are also very specific now adays. So if you wanna be a DevOps, go for it, but if you rather developed apps, don't waste your time on setting up those things from scratch.

Yeah there's the big thing, I'm gonna stick to AWS to show off my skillset there. I don't have some great business idea so I'll get a job until then

Learning aws also gives me cancer but otherwise I agree 👍

Agree 100%. As soon as the video started I immediately pulled up planetscale and saw the DB sizes and immediately closed the page after it appears you can only get a custom size DB by contacting the sales team. 5-10gb ain't gonna cut it for most applications. At least not anything I build. My stuff is data heavy for sure. Almost didn't continue the video but then I checked out Clerk. They have since doubled their free tier to 10k MAU.

like heroku

one major player, not running on VC funding, comes to mind

Doesn’t that assume zero caching?

@@trappar_og Yeah, I was making my capstone project (just to pass school) so no caching yet. It was a CMS app for teachers and students. Is it cachable? I don't know, but I was using 1.24GB of bandwidth with only 5 devs and about 20 users randomly visiting for a month. That's 1/1000 of Vercel's 1TB and I don't know how I feel about that (literally). I'm curious though.

They will message you and ask you to upgrade when you use their service a lot

You could move individual data intensive needs to other services. I can deal with a lot of data via Shipstation so some stuff I do straight through aws s3 then everything possible I do through these free tiers. His point is don’t go crazy getting things setup overboard off the bat.

what was your project if you don't mind? i am also looking for ideas for capstone projects, if you share it i can take some infspiration@@An7Hoang

If you are purely in the stage of educating and perhaps building something yourself, you can get essentially an entire year of free tier micro instance stuff to get yourself going. You might encounter a few dollars a month of cost depending on what you might be using. AWS wants you to be able to learn about their services, as they understandably figure that they will make their money later when you have a viable commercial offering or work for a company that uses aws.

The thing is employers more often use, and look for engineers experienced with, AWS. I've never used these services, they seem awesome, but I'm still in my "get a better job" era so unfortunately will stick with using AWS to build that expertise

Still better than aws rds and aurora 😅

"if you do it right" but how long does it take to learn that?

@@MrMudbill not very long to be honest. amplify is incredibly simple.

@@MrMudbill my cousin with zero cloud experience and just learning dev was able to do it. He watched a youtube video. He barely knew what git was.. no concept of what a CI/CD was and managed to get NextJS + Cognito + dynamo setup pretty easy and free. The declarative security portion tho does take time. But that is a cost you don't keep repaying in regression failures for security.. like specific s3 or dynamoDB access. Some cases eliminate the need for a backend entirely.

I can confirm Free tier for Clerk is for anyone including businesses.

Is there any Theo-blessed product that disallows commercial use on the free tier? I don’t think there is

​@@Mozzius Vercel

@@ItsAvyy Exactly. This is the main one of the stack and it's the biggest reason to use something else for cheaper than the minimum $20 a month.

@@safinn you're building a business product and 20$/month is too much? lmao

I think this is the main selling points for aws or cloud providers. Often larger scale teams the flexibility to deploy an entire environment (dev, test, prod, etc) with a single command. The more third party services you bring to the table the harder it typically is to 100% automate all of this; in fact, some services don’t even provide environments or an api for creating or destroying environments. Additionally, work on any government project in the us and they will state “you must be using aws gov”. But I do see his point, you can get a simple thing deployed in minutes to prod using these tools and you don’t need to waste time with terraform or other complex tooling. I think the goal is make sure these services don’t force vendor locking and have a migration path off the service just in case

That's the thing: with stable (not actually possible in AWS) production you don't need new environments, setting up environments is a first step and highly overhyped. In the end of the day production should work cheaply 24/7/365. So ultimately price of spinning up environment is rather irrelevant. AWS is sort of offering unlimited (fake) scalability and first dose for free but in the end one will find lot of limitations and expenses spiking up for no legitimate reason. Will you ever check why AWS charge you $220 instead of $200? No, it's impossible to track all that small charges that adding up.

Yeah, it seems like his experience is more with lifting and shifting legacy apps. If you design your app for AWS from the start--using Dynamo, Lambda and CloudFront; AWS can be extremely cheap. That said, I may check some of these other services out. Amazon/Bezos already have enough market-share without having to assimilate me, too.

it’s like a 2 minute, 1 step process to set up a budget and then it’s physically impossible to get accidentally charged. As usual the only problems with AWS are between the chair and the keyboard. But you don’t need a certification to know how to do this one!

@@over1498 I know. Doesn't change my stance as that step should not be necessary. It is a deliberate decision by Amazon and is bordering on being a scam.

Why would you add something to your stack before reviewing pricing? That is idiotic

@@programming.jesus1234 Okay, so hear me out. A person I know had a service in the free tier. Said a database of a certain type and amount of traffic was included. Did just that. Turns out there was some bullshit setting in the particular database that ended up with him being charged $200 for around 400 database requests. Not once did AWS warn that it wasn't included in the free tier, and the cost was still 0.

@@CottidaeSEA sorry for your experience dude, but with AWS you really need to spend some energy kinda understanding the budgeting first. They offer several tools for cost management

Thats a rly fair price for that amount of users. What VPS are u using? I'm trying to set up a node project for the fall and don't want to run into insane lightsail pricing.

@@kevdoer2174 Currently I'm still on Vultr, however they are doing some shady stuff now and Im thinking of moving it all to digital ocena.

@@kevdoer2174 Im using Vultr, however, they have some shady TOS update so I think I will go to Digital Ocean..

This is also my biggest concern. When you own everything you can run it for decades without touching it. What happens when your project is a pile of SaaS and you glue them all together with yet another SaaS and one of them fails? Is every service wrapped in an adapter so you can swap it out without a major rewrite?

@@spicynoodle7419 Well in my case I’ve been using Prisma (although I’m feeling the performance hits from it) as an abstraction layer, and was looking at Lucia Auth as an authentication layer to hopefully deal with most of that vendor lock in, but it’s still a bit too much that it keeps me on my toes on trying to find good ways to avoid that. We’ll see how things will go

I would say to not use their APIs directly. Its better if you can abstract them as much as you can.

You avoid them by not using them.

Planetscale is basically just MySQL and if your db is gigantic Vitess in front, both are open source and self hostable although Vitess might not be soo simple, both are industry standards, I think Vercel & especially Clerk are biggest lock-in in this stack, Upstash is basically Redis

if its only one, you practically dont have any redundancy in case this location goes down.

@@dermuschelschluerfer often an accepted risk for dev and staging

@@dermuschelschluerfer We do have another location

@@EraYaN but not for Prod please xD

You have to get to a certain scale for this to make sense. It also very much depends on what your application looks like. It's certainly cheaper to self host, but then you also must self-manage, deal with hardware issues, system administrate, run services, pay for security, and in general reinvent the wheel, and I say this as someone who has been an engineer at companies of all sizes and varieties, including a company that had it's own data center(s) and servers that numbered in the high hundreds. During my regime there I introduced the use of virtualization and pushed the company forward, as standard maintenance, not to mention software releases required a small army of system administration staff, a noc, and 24/7 support.

Yeah, using upstash as a cache or something with just 10k operations / day is really low if you just want a hobby project like this.