@Skyline_NTR before, we're going through that now. Our team is done but we're barely even a fraction of a percent of the cost. 🤷♂️ Our team's few ecs instances, DB, S3, and lambdas only run about $2500 a month for dev/qa and prod.
It's also good that thanks to the European Data Act it's free to transfer out all your data and cloud providers cannot charge outbound traffic for this.
@@SERWERWARMINEPL yeah, EU regulations are actually pretty damn good for consumers and small businesses - and their fines are relative to the revenue of the business, which means companies like apple and google receive fines that might actualy make their share holders wince. I'm not a massive fan of the EU for political reasons, but living in the UK I still wish we hadn't left...
Believe it or not competition requires a decent amount of regulation, otherwise you just concentrate economic power into a few corporate fiefdoms that decide everything.
@@markhausBelieve it or not, regulation leads to overegulation which concentrates power into an actual federalized fiefdom. Free markets > government intervention
I'm convinced the new business growth approach is: 1. Go to cloud immediately. A Lightsail instance is like $5-10/month. There might be a few additional services to setup, then you're done. You focus on the app rather than the hardware. 2. Migrate some services from the cloud to local hardware. Literally just cutting out bloat and cost from your development to your prod. 3. Almost fully migrate off cloud services. Only database off-premises. Absolute control, bills go way down. 4. Hybrid again. Some things need to be connected internationally that just sucks to setup. Leave it to experts. 5. Fully cloud again. You want the ability to blame someone else at this point if anything goes down, which will probably only happen if everyone else also goes down.
We always knew that at a certain size/volume of computing and data it’s more economical to have your own servers. What makes DHH’s cloud exit articles so interesting is that he’s making the case (that thanks to containerization), that point is much closer than most people assumed.
I find managing Linux boxes a lot easier than managing AWS. You have to remember that these days companies hire a lot of people to manage stuff hosted on AWS.
I've never understood it. We've got this system that's open-source, has decades of history and mountains of piled up documentation and tutorials and QA forums, and then you've got a proprietary system where they change all the words to their own internal vocabulary, add a bunch of complexity because of all the billing ins and outs, add a bunch of new and poor quality tooling, and it's supposed to be easier? How was that ever going to be easier?
Suppose that I'm working for a company that doesn't allow us to run docker locally. Now I need a machine that runs a docker daemon so that everybody can use their local docker client to connect to that machine. I think it'll take a couple of days to have that machine because: 1) I have to create a ticket 2) you need to find time to work on that ticket. Then if I want to change the number of cpu, memory on that machine, it'll take a couple of more days. With AWS, it takes me 10 mins.
I think you said it before. Starting out, the cloud is great. It scales and even more important, is redundant. You don't have to deal with "the cpu in our server is dead and it takes X hours to replace it". If you are small and can only afford one server or two under powered servers then you can be down and that costs. There comes a point where you are big enough to have enough redundancy due to your size. That's when going onprem becomes a viable option. Maybe not right away, but that would be the earliest.
It is more of a way to rent yourself out of a bind or a whole area of tech/knowledge debt. For a small, inexperienced team it works better to focus on the skills needed for immediate application, but at some point it becomes smarted to develop those. Or, in the case that someone already has the skill/knowledge, it is very inexpensive to start and keep things in house.
The name... is the not ready to hitchhike across the galaxy...agen. Really interesting updated from DHH on how their decoupling from the cloud is going. It's strange that you find people putting nasty labels to that process rather than just following the engineering mind frame by asking what's been learned and having retrospectives on the process. Great content as always.
I mean, economy of scale _does_ scale here, it's just that the guys selling the stuff know they can get away with selling inverse economies of scale (and if you think about it, opaque systems like this do benefit from that sales model), so they do. They wouldn't be able to get away with it if the C-suite had a clue, though. Education and transparency are the enemy of this sort of sales model. Personally, I'm kinda glad it works this way: large companies who can afford to pay stupid prices do so, and subsidize the little guy just trying to make a startup. Capitalism at its finest, I say.
It does scale, it's just that there mr Jeff Bezos capturing it by being the middle man. Own your servers then you can benefit from the economies of scale. Modern servers are amazingly powerful. And you don't need to go back to the insanity that was on-prem in 2010. You use open source private cloud and just manage your fleet of servers like a cloud provider. Totally automated, you can even automate server failure and replacement using the remote hands of your co-location when failures happen. You can have all the good benefits of having cloud software and devops without the absurd OPEX just for a small CAPEX price that you recoup in 6 months of operation. The only moment you don't go for on prem and go for cloud is if you have bombastic growth, but thats a good problem to have. People want to be like the Netflix and create all sorts of insanity with microservices and kubernetes and FaaS, when they needed for their business was a monolith backend app and 2 or 4 servers.
I'm working for a corporation and we're moving to the cloud, from an on-prem K8S solution. We need it because we keep having scaling issues - the amount of people trying to run more and more keeps increasing, and we can't drag hardware fast enough into the building, so The Cloud ™️ is a good solution for us, for now. Can't wait for 2 to 4 years down the road when corporate finds out how expensive it all is 😅
Could be that your software needs optimizations too. While there is a limit on what gains you can get, for most companies there is a ton of low hanging fruit.
Your coders should write code to use a hybrid infra. Keep increasing the on-prem hardware to fit the base workload. Only scaling goes to cloud. If your coders are writing inefficient code, then all and any hardware will fail for sure.
iAAs (infrastructure as a service) - which can be done on prem with systems like open stack with terraform and sufficiently sized modern hardware has great and always improving economics - amd has had 8TB ram systems for years with significant core counts and especially or alternately ARM64 hardware lowers power costs - network rates can be in the 200gbps range between very dense hosts with extremely good storage access rates to Nvme flash arrays Data has gravity and the more localized data is with compute the less optimization and other work on distributed systems is required. If raspberry pi’s were more available and SSD and cpu production hadn’t been limited by lack of Taiwanese fabs everyone would have a petabyte data center sized system at home enjoying massively faster rates than cloud systems - but perhaps with backups being difficult as FTTH rates are held down by lack of infrastructure maintenance spending
@@tomorrow6 Important caveat that you left out. These things can be done on-prem if you have/can retain the required talent. That's not likely to be true for many or maybe even most enterprises.
I love my BareMetal's, I love this awesome open source community that create those huge tools that allow us to scape from the cloud and create our own!
Isn't it a simple math question? If you have another company doing your hosting, they do wanna make a profit off of it. So as long as your company is big enough to run your own datacenter, it's always gonna be cheaper to do it yourself since you cut out another profit oriented party.
The math part you refer to is typically easy to see. The hardest part about actually convincing a company is that some companies may not even have their accounting set up to compare the two operational expenses (cloud vs on-prem).
Very unclear. A lot of the functionality of cloud isn't based on just running servers, it's on value-add services that have seen combined decades of work. Even with servers, the value of what you're selling and your reliability needs matter. If your on-prem offering gives 99.9% uptime but cloud providers are giving you 4, that .09% can actually be net revenue positive. Especially when you include the operational cost of giving many teams that 3 9 guarantee. If you're just running servers and don't have strong uptime requirements, then yeah, maybe moving all of your stuff to a colo or a budget provider makes sense.
Datacenter is not what you really need. You need datacentERS - many of them. All over the globe: at several places in US, some servers in Europe etc. Even if you limit your business to just US, you still better have several centers here and there. US is big.
I think the "optimal" choice wavers back and forth depending on the scale of your product and/or company. That said, people vastly over-estimate the failure rate of hardware. Most of ours is hard drives, which you can get remote hands to replace. Other than that, our most common failure is capacitors on motherboards bursting. We did have a bad batch of power supplies that tended to shoot flames out the back. Most of our servers are over 10 years old. We have one that is 22.
There are solutions for hybrid clouds where you can move processes between your on-prem servers and the cloud. So you can mix and match and get the best of both technologies. But that's only for real engineers that don't just go for what's popular and choose things based on their use case without caring what others are doing. This profession has this problem where people literally follows others jumping to the pits and jump too, it's insane. I guess it's like pop music, ridiculous.
I replace the old capacitors with new ones once the mobo is > 5 years. And I am absolutely adamant on using the best PSU so that the hardware lasts long. Even their caps are replaced after 5 years to keep power clean. If you are using 10y old servers then you are bound to have problems, No surprises.
I started with cluster on prem from the very beginning because I knew I will be cost limited (and I'm the only dev). It allowed me to develop a software that would require at least $1-2k of costs per month - with current costs of around $20 per month. Initial investment in hardware would be burnt in 3-4 months on cloud. I'm not counting my time I've spent in setting it all up, but after 2 years it was definitely worth it.
@@HerbieBancock I'm not a native speaker, so that is where it comes from. I read through some nice explanations as to why it's considered incorrect and will try to avoid repeating it in the future, thanks!
The cloud usually gets you on the relational database, Cloud relational databases are roughly 10x as expensive as relational databases outside of the cloud.
I wish I could read Wheel of Time for the first time again. There's lots to see in a reread that you didn't notice first time around, but there's nothing quite like experiencing such an expansive setting for the first time.
Wow you are joining at just the time that AWS bumps prices UP so it can fund more AI LLM compute (depending - AWS is also saving money from traditional compute moores law improvements like ARM64 based compute that historically Fortune 500 companies have benefited from) - compute has increased as more applications have become available and that ain’t stopping
"We have been lied over here?" R: Yes. The quantitiy of propaganda promoting this cargo culting is beyond ridiculous. We can do things WAY easier and cheaper.
I also work for a very large service company that is moving back to on prem, and the problem that we’re running into is having to mass hire developers in order to build out the functionality that will be lost. I have no idea how much will be saved versus how much is being spent in hardware, maintenance, personnel, support, etc. But we are definitely not doing it with the same number of employees that we had. My guess is that your costs will increase in areas that you aren’t expecting or even noticing. For instance, spinning up a dozen new VMs and/or clusters a day means that those servers have to be up and running in advance, which means that you have to have more hardware and more people than you actually need, or else productivity stops, which is both a waste of money and something you don’t have to worry about in the cloud. As an analogy, you can save a lot of money doing your own yard work, painting your own house, fixing your own plumbing, doing your own finances, managing your own investments, etc., but at some point, there are no longer enough hours in a day and something is going to have to be put off until later. And putting something off is also a hidden cost. But I do think that everyone moving away from the cloud is a good thing, because it will force the providers to cut their prices to the bare minimum in order to stay relevant, and at that point, we can move everything back to the cloud, because it will be cheaper. 👍
Lots of folks here are talking on-premise vs cloud and managing physical hardware. That’s not what DHH and Basecamp are doing. They basically moved to colo-as-a-service. They still aren’t touching drives or cables. Deft runs all that for them. It’s pretty close to IaaS but on their hardware. Big +1 from me.
You're one of the people that got me into WoT! Book 11! I'm halfway through 8 right now, but I've got the rest on the shelf behind me waiting for their turn. Surely it'll only be a few months... Are you still reading them to your kids? I feel like I remember you telling us about that a few months ago.
There is a video outhere tittled "the cloud is over engineered" Btw, I read some comments of people saying that after lowering the AWS bill they got laid off anyway...
Always remember this when executives want you to work free overtime because it is critical for the company. Those are the same folks who will be thrilled to lay you off.
Pure Storage is so good. I don't know why ppl don't use it. As DBA, for big databases is almost a must. Cloud is old school trickery, $1 to enter $100 to exit. Everybody was thinking "Oh, I am in cloud I don't need to pay Engineers anymore" ... yeah, but you have to pay DevOps Engineers :)
Really it's more about memory consumption more than speed. But again, as usual, most people don't use their spare performance to run more, but to run more shitty code for longer.
The biggest performance bottleneck is almost always the DB. Meanwhile, 37signals rakes in millions with actually selling products using a stack they own and know and are effective in. Customers don’t care about tech stack.
@@avwie132 It doesn't matter if the bottleneck is the DB. The point is that they still saved a ton, even if they use one of the most inefficient stacks. No need to run defence lol
@@sarabwt I am not getting defensive. But what point are you trying to make? I'd imagine that having an inefficient stack is one of the reasons you save a ton.
@@avwie132 I'm making a point that cloud gets expensive very quick and that it almost never makes sense to be 100% in it. You can "overprovision" hardware and it would still cost you way less than the optimized resources in the cloud.
As a devops engineer, i will gladly drain my company's bank account to never have to touch hardware or deal with dying drives and that crap ever again. Cloud is worth whatever it costs
If it wasn't worth the cost, aws would be cheaper. Theyre not monopoly, just have much better product than what you get on perm. Security is the most important value of cloud.
@@adriankal Per AWS "While AWS manages security of the cloud, you are responsible for security in the cloud." Its not nothing, but it also isn't a lot.
Every place I have worked went to the cloud thinking they would be saving so much. Really, they just like outsourcing because it looks simple on a spreadsheet at a board meeting. It makes everything more tedious for the core developers.
I had clients that went from on prem to IAAS to on prem in less then 10y and some going from IAAS to on prem. All of them saved money going on prem after a short period of time. But I also know a few clients that have on prem and cloud services and they save money with that construct. Also AWS is in "public hands" since Amazon is on the stock market, anybody can own parts of Amazon...
The all in costs for labor are pretty insane when it's all inclusive. I work for a major tech company, and my pay is only 30% of what it costs for my role
And then realize how much employee turnover costs. You've got the loaded salary for a year getting up to speed plus all the opportunity cost waiting. Crazy how many companies think this is fine.
Hybrid is always the way to go. The cloud makes sense for 'new' tech or to improve all tech on a large scale. But just like mainframes it makes more sense to go back to more 'localalized' versions onprem. Not with subscriptions but with ownerships, subsriptions only make sense in the cloud where there is a lot of operational cost. And this can go back and forth over time to get a hybrid situation for things that can go either way, in the cloud using subscriptions, on-prem using perpetual licenses and of course the hybrid case. Cloud first or on-prem first is not the solution, it's always a combination of both where it makes sense for a business.
The insidious thing about the cloud is that as it reduces the use of local server infrastructure, it also reduces the amount of people skilled in the area, which in turn increases the cost of jumping off cloud
IMO if you have a successful product with reliable customers, a usage pattern w/o extreme peaks, and someone who knows what they're doing, on-prem makes much more financial sense. Not every company meets those criteria, and that's perfectly ok. But for those that do, building on Kubernetes opens the door of possibilities with tons of really good open source projects, and you can likely reach a density and cost advantage thats just not possible with the cloud tax. I think Kubernetes simplifies on-prem operations more than it complicates things. My current job is moving in this direction, and even at small scale the numbers just make more sense.
Meanwhile, people complaining here in the comments that Ruby is slow. While they’re reinventing the wheel for the 115th time in the newest next gen SSR JS Blazingly Fast Framework, only to notice that in real projects it is the DB that is the bottleneck.
All the complaints about having to replace a broken server would make sense if cloud costs were really controlable. I’d rather roll the dice on having to buy new hardware than on AWS sending me a bill with 1000x of my typical spending.
So long and thanks for all the fish is actually The Hitchhicker's Guide To The Galaxy reference. I think that was the title of the 4th book in the series. Brilliant British humour. The source of the "Answer to the great question of the universe, life and everything" bit.
Colocation is not "on-prem". Colocation means renting rackspace, power, IP transit and NOC personnel from an existing data center (similar to dedicated bare-metal servers, but you own the hardware). On-premises literally means doing everything yourself in your own building.
Worked at at large saas. We did a cost savings push on Amazon and after 3 projects we saved 15mil a month. We did more aggressive peering over transit gateway, swapped to arm, and moved to karpender.
I joined a company that leans HARD into AWS, not just cloud but all the AWS specific primitives, Lambda, SQS, SNS, EventBridge, etc. It seemed a little coolaidy at first but actually it's been great. Our traffic is VERY bursty, we have two massive spikes every day and then basically zero, and a fraction of that over summer. We need super high uptime and reliability. The application is very interlinked and complex. Probably 80-90% of CRUD applications that just have large storage needs don't actually need cloud in all likelihood, but in some cases it make a ton of sense.
Found a server rack outside in NYC, paper said on it that it works, guy bought it on eBay for $350, wrote that it can be sold $200 easily, I just took it and keep it at home, but have no time to play around it
I hate that modern software engineering culture has become about all or nothing opinions. Either cloud good + on-prem is the devil, or cloud bad + on-prem is the only way. Just solve problems with the right solution. Cloud is the right solution sometimes. On-prem is the right solution other times. Cloud empowers small and early stage teams/developers. Without cloud computing, so many businesses would be dead before they even had a chance. At the same time, it is expensive and sacrifices lots of control. Decide whats right for you based on the facts of your circumstances, don't make your decisions based off of internet discourse, and what some other guy did.
I was for a tjme lead architect for infrastructure at a large online game company who operated in the cloud (not Zynga, the other big one with actual games games), and there had been a preliminary decision to leave the cloud just before my hire. I was asked to analyze and build a business case either way. Buying hardware, all the premium datacenter costs, the additional maintenance was cashflow neutral end of first year and saved $20 million total expenditures over 36 months (+-25% depending on business growth assumptions and how you applied capex and opex, which the CFO hadn’t decided a standard for yet). The $7.5 million hardware outlay was the easiest infrastructure check I ever got signed. Only glitch was that we ordered right as an industry wide SSD generation shift in production hit, and we were six weeks late due to waiting for Intel’s first thousand units early customer access to ship (only half ours). Cost savings overall hit the numbers.
I've been saying this for years. Cloud is more expensive in the long run once you move past just having a few services. I am a big believer in trying to keep as much stuff onprem as possible to keep costs low and you don't have to spend all the time moving back and forth.
For now I'm cloud based. When your operating expenses for AWS are still under $100 a month with over $150K ARR, you're tiny. But there is no way going to self-hosted would be cost or time effective at this point. Not when you're the only person (4 hours east of ThePrimeagen) in the middle of corn fields.
With the hitchhiker guide I believe there’s a complete video of the author reading it on UA-cam. Really enjoy the authors intonation of the characters. Found it more enjoyable than the movie if I would be honest.
You have other options, you can move to a COLO... Reminder, if you on Cloud, you still have infrastructure engineers. Cloud is just expensive.. I work for a company that uses AWS, Google and Microsoft azure. you talked about S3, you can do that on prem.
I saw the same story in another company that I worked for as a contractor. Basically no cloud = huge savings. Cloud is probably only good to start or as a backup infra. If you start in cloud and grow enough, then the time will come to leave it.
I specialize in Cloud cost optimization (FinOps). It’s crazy to see how many companies are not utilizing the cost optimization levers in the cloud. Often because of a lack of responsibility for the costs.
I recommend listening to The Hitchhiker's Guide to the Galaxy because Douglas Adams originally wrote it as a radio show. It sounds a bit better than it reads.
My team of 8 people including an EM manages ~ 6000 VMs and 3500 macminis hosted on company owned and third party DCs and we save a ton by not hosting on cloud. If your work load is stable it pays to be not on cloud, ops would be high but savings is still substantial
it does not make sense because having your own physical servers is always more expensieve, power, admin sys, more people and contractors. It is always cheap when your business is on apps and websites rather than the whole operations.
Government not seizing but instead heavily regulating that kind of business is a common take in most countries including the US since when a business adopts a monopolistic position on a market kinda like it does in the electric generation and distribution sector its often heavily regulated and striped from its "supper-normal profits" (referring characteristic of a monopolistic market in microeconomics) so the price is equal to the marginal cost of that good or service. Its not communism, its pragmatism. The US, Canada, France, UK, and many more countries do that with for example electricity generation and other naturally monopolistic sectors. On the other hand it could be feasible that the market ends up regulating itself (if that were the case we could be certain that its not a monopoly). It would be caused by people just leaving the cloud and going to on premise, or instead just by the different cloud providers competing against each other.
Managing 3 devs is not a full time job, unless you seriously hired the wrong people. Unless you’re a small startup, you probably have an existing manger or even the CTO who can handle an extra couple of meetings per week plus whatever review process and admin busywork for 3 more staff. It would be unlikely you’d hire a new manager for such a small team. Usually they farm those duties out to a team lead which is an awful role but getting popular due to the cost cutting of making a senior dev pull triple duty as a dev, manager, and usually a PM as well.
They usually totally forget about all of the labor & taces and ancillary costs like insurance, power, redundant fiber uplink, etc. The hardware is the cheapest part of the equation.
I thought everyone knew that cloud doesn't save money, it saves time. My company is trying to move from On-Prem to cloud and we have spent a year spending 10x the quoted cost from AWS and the performance is still worse.
The Cloud hype is the new Microsservices hype. Hope people realize finally it's not a silver bullet. It has tradeoffs and it's not always worth it. Never underestimate private services costs and architectural complexity
9:55 I mean I get why they hate DHH as I did too originally but the more I saw of his takes the more of it seemed fine. There are still many takes I disagree with but most of the takes that get surfaced are ones I would disagree with in the first place. Those opinions you never hear of are the one you will agree with. I can't hate someone for having takes I disagree with, at least not when it's not the majority of the takes.
1:55 I find this to be an out dated view of the company hierarchy. You're correct that this is how it would have been done traditionally. However I would say in general if employees are professionals such as engineers then there is less of a requirement for oversight. Instead you can have a lead professional do the necessary required work to facilitate the business function all whilst carrying out the same work as the other non lead professionals. Any specific employment issues would be dealt by HR as they most likely would've been if a manager was involved. The advantage of having a lead that also does the job is that they'll also understand the issues on the "shop floor" and may also be able to be proactive in resolving them.
AWS cloud architect, most of the time on prem is probably what you need. Large conpanies that have a lot of red tape benefit from the cloud as their change process relies more heavily on third party vendors which is actually a very large body of work. Bursty workloads do better in the cloud but the vast majority of applications arent actually that bursty. With K8s its much easier to lift and shift things around to other providers to get the best bang for your buck.
With a 100% of certainty, Those not knowing linux in software world with no hardware perspective are not only Jnior devs but also weak devs and forever be if continued without learning.
Don't use IaaS unless you have to. There is a tipping point where the volume is so high that IaaS clusters could be more cost-effective than serverless solutions like Lambda, State Machines, and Fargate containers. However, most organizations will not reach that tipping point. When people leave machines running 24x7 under-utilized, you can expect a nice fat bill. Also entrust it to people who are well aware of costing strategies and which services to use.
I hope everyone doesn’t have short term memory of what AWS did to Parlor. It was likely illegal but since their politics aligned with the ruling class they got a pass. Unless forced I will NEVER use AWS. One day they could 🤬 you over too.
Thinking not using cloud services is “thinking different”. Translation: I’ve been doing for less than 10 years and think IT didn’t exist before I learned about it.
I work for a large company. Their cloud bill is 2.5 million PER MONTH. We aren't even a tech company. We're a manufacturing company. It's kinda nuts.
is that before or after cloud spend optimization
@Skyline_NTR before, we're going through that now. Our team is done but we're barely even a fraction of a percent of the cost. 🤷♂️
Our team's few ecs instances, DB, S3, and lambdas only run about $2500 a month for dev/qa and prod.
@@Skyline_NTR After. Before the optimization it was 1 million.
@@Skyline_NTR what is cloud spend optimization?
Moving the idle-compute slider down from 10 minutes to 1.
It's also good that thanks to the European Data Act it's free to transfer out all your data and cloud providers cannot charge outbound traffic for this.
wtf I love eu now
@@SERWERWARMINEPL yeah, EU regulations are actually pretty damn good for consumers and small businesses - and their fines are relative to the revenue of the business, which means companies like apple and google receive fines that might actualy make their share holders wince. I'm not a massive fan of the EU for political reasons, but living in the UK I still wish we hadn't left...
AHH, so that's why pricing changed, thanks that explains it. Very good of them and they are right, it was anti-competitive
Believe it or not competition requires a decent amount of regulation, otherwise you just concentrate economic power into a few corporate fiefdoms that decide everything.
@@markhausBelieve it or not, regulation leads to overegulation which concentrates power into an actual federalized fiefdom. Free markets > government intervention
I'm convinced the new business growth approach is:
1. Go to cloud immediately. A Lightsail instance is like $5-10/month. There might be a few additional services to setup, then you're done. You focus on the app rather than the hardware.
2. Migrate some services from the cloud to local hardware. Literally just cutting out bloat and cost from your development to your prod.
3. Almost fully migrate off cloud services. Only database off-premises. Absolute control, bills go way down.
4. Hybrid again. Some things need to be connected internationally that just sucks to setup. Leave it to experts.
5. Fully cloud again. You want the ability to blame someone else at this point if anything goes down, which will probably only happen if everyone else also goes down.
Or you can use unikernels in the cloud and don't give a fuck about your bill
This is a good starter blueprint.
About the third one, keeping database and application server away from each other could be an issue.
Like so many things, whatever the current thing is will go too far and cycle back with a massive over correction.
@@batuhanaydn4592 I should have been clearer that I meant you have a backup off-prem "in the cloud."
The name is the "On-Premagen"
The name......
RansomMagen
We always knew that at a certain size/volume of computing and data it’s more economical to have your own servers. What makes DHH’s cloud exit articles so interesting is that he’s making the case (that thanks to containerization), that point is much closer than most people assumed.
I find managing Linux boxes a lot easier than managing AWS.
You have to remember that these days companies hire a lot of people to manage stuff hosted on AWS.
I've never understood it. We've got this system that's open-source, has decades of history and mountains of piled up documentation and tutorials and QA forums, and then you've got a proprietary system where they change all the words to their own internal vocabulary, add a bunch of complexity because of all the billing ins and outs, add a bunch of new and poor quality tooling, and it's supposed to be easier? How was that ever going to be easier?
Seriously AWS is a kafkaesque nightmare, how are so many people scared of a Linux ssh session by comparison
Suppose that I'm working for a company that doesn't allow us to run docker locally. Now I need a machine that runs a docker daemon so that everybody can use their local docker client to connect to that machine. I think it'll take a couple of days to have that machine because: 1) I have to create a ticket 2) you need to find time to work on that ticket. Then if I want to change the number of cpu, memory on that machine, it'll take a couple of more days. With AWS, it takes me 10 mins.
@@avalagum7957 And $50,000 because you forgot to properly clean up the docker instances
@@avalagum7957but you're complaining about your company policies. Nothing prevents the company from automatic provision of whatever you need.
I think you said it before. Starting out, the cloud is great. It scales and even more important, is redundant. You don't have to deal with "the cpu in our server is dead and it takes X hours to replace it". If you are small and can only afford one server or two under powered servers then you can be down and that costs.
There comes a point where you are big enough to have enough redundancy due to your size. That's when going onprem becomes a viable option. Maybe not right away, but that would be the earliest.
It is more of a way to rent yourself out of a bind or a whole area of tech/knowledge debt. For a small, inexperienced team it works better to focus on the skills needed for immediate application, but at some point it becomes smarted to develop those. Or, in the case that someone already has the skill/knowledge, it is very inexpensive to start and keep things in house.
The name... is the not ready to hitchhike across the galaxy...agen.
Really interesting updated from DHH on how their decoupling from the cloud is going. It's strange that you find people putting nasty labels to that process rather than just following the engineering mind frame by asking what's been learned and having retrospectives on the process. Great content as always.
"So long and thanks for all the fish!" is a reference to Hitchhiker's Guide to the Galaxy. ;3
Can't wait for Theo to try and make some argument for why this is bad and why cloud is actually the savior after all.
not the cloud but vercel. Because, we should we doing embedded systems with Next.js
@@roccociccone597 i prefer to wait.
Wish the argument will include SQL queries inside of frontend code
Upload thing that he developed stopped using serverless and it saw like a 5x speed up in upload time.
@@ithinkimhipster502 shocking no one expected that
A weird world we live in where economy of scale doesn't scale.
It scales exactly as designed, for aws
@@adambickford8720 for customers it still scales. Compare how much you can save with AWS compared to something like Vercel... Crazy.
I mean, economy of scale _does_ scale here, it's just that the guys selling the stuff know they can get away with selling inverse economies of scale (and if you think about it, opaque systems like this do benefit from that sales model), so they do.
They wouldn't be able to get away with it if the C-suite had a clue, though. Education and transparency are the enemy of this sort of sales model. Personally, I'm kinda glad it works this way: large companies who can afford to pay stupid prices do so, and subsidize the little guy just trying to make a startup. Capitalism at its finest, I say.
@@mage3690pretty much - the free tiers small shops develop on (and rotate through quickly) depend on the spending of the larger customers cloud bills.
It does scale, it's just that there mr Jeff Bezos capturing it by being the middle man.
Own your servers then you can benefit from the economies of scale.
Modern servers are amazingly powerful.
And you don't need to go back to the insanity that was on-prem in 2010. You use open source private cloud and just manage your fleet of servers like a cloud provider.
Totally automated, you can even automate server failure and replacement using the remote hands of your co-location when failures happen.
You can have all the good benefits of having cloud software and devops without the absurd OPEX just for a small CAPEX price that you recoup in 6 months of operation.
The only moment you don't go for on prem and go for cloud is if you have bombastic growth, but thats a good problem to have.
People want to be like the Netflix and create all sorts of insanity with microservices and kubernetes and FaaS, when they needed for their business was a monolith backend app and 2 or 4 servers.
In my 900 person company, we pay Jeff Bezos 12M a year for AWS and we also have highly paid MBA program managers to keep the cloud costs down 😂
😂 This means, Jeff Bezos has better MBA guys to fool your MBAs.
....JFC , 13+K PER employee.
I'm working for a corporation and we're moving to the cloud, from an on-prem K8S solution. We need it because we keep having scaling issues - the amount of people trying to run more and more keeps increasing, and we can't drag hardware fast enough into the building, so The Cloud ™️ is a good solution for us, for now. Can't wait for 2 to 4 years down the road when corporate finds out how expensive it all is 😅
consider VPS providers as well, usually more cost effective than AWS & co
well its better to scale quickly and capture the market share, then after things level out move back out of the cloud. at least I would think
Could be that your software needs optimizations too. While there is a limit on what gains you can get, for most companies there is a ton of low hanging fruit.
We moved from a rack in which we owned the hardware to AWS. Everyone freaked when our costs went from 10k / month to 50k/month.
Your coders should write code to use a hybrid infra. Keep increasing the on-prem hardware to fit the base workload. Only scaling goes to cloud.
If your coders are writing inefficient code, then all and any hardware will fail for sure.
When you begin to view “cloud” as software-defined data centers, it’s easier to see the pros/cons of why someone might want to move back to on-prem.
iAAs (infrastructure as a service) - which can be done on prem with systems like open stack with terraform and sufficiently sized modern hardware has great and always improving economics - amd has had 8TB ram systems for years with significant core counts and especially or alternately ARM64 hardware lowers power costs - network rates can be in the 200gbps range between very dense hosts with extremely good storage access rates to Nvme flash arrays
Data has gravity and the more localized data is with compute the less optimization and other work on distributed systems is required.
If raspberry pi’s were more available and SSD and cpu production hadn’t been limited by lack of Taiwanese fabs everyone would have a petabyte data center sized system at home enjoying massively faster rates than cloud systems - but perhaps with backups being difficult as FTTH rates are held down by lack of infrastructure maintenance spending
Also the whole AI push is really favoring companies with existing data centres and retained long term low power cost contracts with power vendors
And you can run something like Ubuntu MAAS to get a working API
@@tomorrow6 Important caveat that you left out. These things can be done on-prem if you have/can retain the required talent. That's not likely to be true for many or maybe even most enterprises.
I love my BareMetal's, I love this awesome open source community that create those huge tools that allow us to scape from the cloud and create our own!
Shout out to reddit's self-hosted subreddit! It's been a goldmine of information!
Isn't it a simple math question? If you have another company doing your hosting, they do wanna make a profit off of it. So as long as your company is big enough to run your own datacenter, it's always gonna be cheaper to do it yourself since you cut out another profit oriented party.
The math part you refer to is typically easy to see. The hardest part about actually convincing a company is that some companies may not even have their accounting set up to compare the two operational expenses (cloud vs on-prem).
not always true because you don't get the economy of scale
Very unclear. A lot of the functionality of cloud isn't based on just running servers, it's on value-add services that have seen combined decades of work.
Even with servers, the value of what you're selling and your reliability needs matter. If your on-prem offering gives 99.9% uptime but cloud providers are giving you 4, that .09% can actually be net revenue positive. Especially when you include the operational cost of giving many teams that 3 9 guarantee.
If you're just running servers and don't have strong uptime requirements, then yeah, maybe moving all of your stuff to a colo or a budget provider makes sense.
Ever heard of economies of scale? No - it doesn't work like that. Quite the opposite.
Datacenter is not what you really need. You need datacentERS - many of them. All over the globe: at several places in US, some servers in Europe etc. Even if you limit your business to just US, you still better have several centers here and there. US is big.
I think the "optimal" choice wavers back and forth depending on the scale of your product and/or company.
That said, people vastly over-estimate the failure rate of hardware. Most of ours is hard drives, which you can get remote hands to replace. Other than that, our most common failure is capacitors on motherboards bursting. We did have a bad batch of power supplies that tended to shoot flames out the back.
Most of our servers are over 10 years old. We have one that is 22.
There are solutions for hybrid clouds where you can move processes between your on-prem servers and the cloud. So you can mix and match and get the best of both technologies.
But that's only for real engineers that don't just go for what's popular and choose things based on their use case without caring what others are doing.
This profession has this problem where people literally follows others jumping to the pits and jump too, it's insane. I guess it's like pop music, ridiculous.
I replace the old capacitors with new ones once the mobo is > 5 years. And I am absolutely adamant on using the best PSU so that the hardware lasts long. Even their caps are replaced after 5 years to keep power clean. If you are using 10y old servers then you are bound to have problems, No surprises.
This is a very good point. The evergreening of cloud vs nevergreen of on-prem. Nevergreening due to org. inertia.
10:07 "They are using one single machine". Wait until he find out a bazillion EC2 "machines" are running in a single machine
1:30 "a lot of people don't understand stuff" You said it, brother.
I started with cluster on prem from the very beginning because I knew I will be cost limited (and I'm the only dev). It allowed me to develop a software that would require at least $1-2k of costs per month - with current costs of around $20 per month. Initial investment in hardware would be burnt in 3-4 months on cloud. I'm not counting my time I've spent in setting it all up, but after 2 years it was definitely worth it.
Do you also say "a hardware" in your country?
@@HerbieBancock I'm not a native speaker, so that is where it comes from. I read through some nice explanations as to why it's considered incorrect and will try to avoid repeating it in the future, thanks!
I too went the same way. I now have 2 racks. More than cost effectiveness, its about freedom of implementation.
The cloud usually gets you on the relational database, Cloud relational databases are roughly 10x as expensive as relational databases outside of the cloud.
"So long and thanks for all the fish" - Douglas Adams, Hitchhikers Guide to the Galaxy
I wish I could read Wheel of Time for the first time again. There's lots to see in a reread that you didn't notice first time around, but there's nothing quite like experiencing such an expansive setting for the first time.
I work for a fortune 25 company. We have 3 gigantic data centers. We moved 1 to AWS and 2 others on the way. We are 100% committed to cloud
congratulations. and?
@ cloud is cheaper and more flexible. Thank you
Wow you are joining at just the time that AWS bumps prices UP so it can fund more AI LLM compute (depending - AWS is also saving money from traditional compute moores law improvements like ARM64 based compute that historically Fortune 500 companies have benefited from) - compute has increased as more applications have become available and that ain’t stopping
Using AWS? You must be idiots
Who fucking cares?
"We have been lied over here?" R: Yes. The quantitiy of propaganda promoting this cargo culting is beyond ridiculous. We can do things WAY easier and cheaper.
At this point, the more I see/hear people promoting something in tech, the more I assume it's ultimately bullshit.
There is the certification aspect of cloud providers that a lot of companies leverage in security audits and such.
I also work for a very large service company that is moving back to on prem, and the problem that we’re running into is having to mass hire developers in order to build out the functionality that will be lost. I have no idea how much will be saved versus how much is being spent in hardware, maintenance, personnel, support, etc. But we are definitely not doing it with the same number of employees that we had. My guess is that your costs will increase in areas that you aren’t expecting or even noticing.
For instance, spinning up a dozen new VMs and/or clusters a day means that those servers have to be up and running in advance, which means that you have to have more hardware and more people than you actually need, or else productivity stops, which is both a waste of money and something you don’t have to worry about in the cloud.
As an analogy, you can save a lot of money doing your own yard work, painting your own house, fixing your own plumbing, doing your own finances, managing your own investments, etc., but at some point, there are no longer enough hours in a day and something is going to have to be put off until later. And putting something off is also a hidden cost.
But I do think that everyone moving away from the cloud is a good thing, because it will force the providers to cut their prices to the bare minimum in order to stay relevant, and at that point, we can move everything back to the cloud, because it will be cheaper. 👍
Lots of folks here are talking on-premise vs cloud and managing physical hardware. That’s not what DHH and Basecamp are doing. They basically moved to colo-as-a-service. They still aren’t touching drives or cables. Deft runs all that for them. It’s pretty close to IaaS but on their hardware. Big +1 from me.
True, the future is colocation.
You're one of the people that got me into WoT!
Book 11! I'm halfway through 8 right now, but I've got the rest on the shelf behind me waiting for their turn.
Surely it'll only be a few months...
Are you still reading them to your kids? I feel like I remember you telling us about that a few months ago.
There is a video outhere tittled "the cloud is over engineered"
Btw, I read some comments of people saying that after lowering the AWS bill they got laid off anyway...
Always remember this when executives want you to work free overtime because it is critical for the company. Those are the same folks who will be thrilled to lay you off.
Pure Storage is so good. I don't know why ppl don't use it. As DBA, for big databases is almost a must. Cloud is old school trickery, $1 to enter $100 to exit. Everybody was thinking "Oh, I am in cloud I don't need to pay Engineers anymore" ... yeah, but you have to pay DevOps Engineers :)
Also, to put it into perspective, these guys are using Ruby, one of the worst stacks in terms of performance lol.
Really it's more about memory consumption more than speed.
But again, as usual, most people don't use their spare performance to run more, but to run more shitty code for longer.
The biggest performance bottleneck is almost always the DB.
Meanwhile, 37signals rakes in millions with actually selling products using a stack they own and know and are effective in. Customers don’t care about tech stack.
@@avwie132 It doesn't matter if the bottleneck is the DB. The point is that they still saved a ton, even if they use one of the most inefficient stacks. No need to run defence lol
@@sarabwt I am not getting defensive. But what point are you trying to make? I'd imagine that having an inefficient stack is one of the reasons you save a ton.
@@avwie132 I'm making a point that cloud gets expensive very quick and that it almost never makes sense to be 100% in it. You can "overprovision" hardware and it would still cost you way less than the optimized resources in the cloud.
As a devops engineer, i will gladly drain my company's bank account to never have to touch hardware or deal with dying drives and that crap ever again. Cloud is worth whatever it costs
Sounds like you don't like devops. Maybe do development and not operations.
If it wasn't worth the cost, aws would be cheaper. Theyre not monopoly, just have much better product than what you get on perm. Security is the most important value of cloud.
@@adriankal Per AWS "While AWS manages security of the cloud, you are responsible for security in the cloud." Its not nothing, but it also isn't a lot.
Every place I have worked went to the cloud thinking they would be saving so much. Really, they just like outsourcing because it looks simple on a spreadsheet at a board meeting. It makes everything more tedious for the core developers.
I had clients that went from on prem to IAAS to on prem in less then 10y and some going from IAAS to on prem. All of them saved money going on prem after a short period of time. But I also know a few clients that have on prem and cloud services and they save money with that construct. Also AWS is in "public hands" since Amazon is on the stock market, anybody can own parts of Amazon...
The all in costs for labor are pretty insane when it's all inclusive.
I work for a major tech company, and my pay is only 30% of what it costs for my role
And then realize how much employee turnover costs. You've got the loaded salary for a year getting up to speed plus all the opportunity cost waiting. Crazy how many companies think this is fine.
Hybrid is always the way to go. The cloud makes sense for 'new' tech or to improve all tech on a large scale. But just like mainframes it makes more sense to go back to more 'localalized' versions onprem. Not with subscriptions but with ownerships, subsriptions only make sense in the cloud where there is a lot of operational cost. And this can go back and forth over time to get a hybrid situation for things that can go either way, in the cloud using subscriptions, on-prem using perpetual licenses and of course the hybrid case. Cloud first or on-prem first is not the solution, it's always a combination of both where it makes sense for a business.
I only use the cloud as a fail over to my bare metal.
The insidious thing about the cloud is that as it reduces the use of local server infrastructure, it also reduces the amount of people skilled in the area, which in turn increases the cost of jumping off cloud
There is already a ton of turnover in the field, this might make things worse but knowledge loss is already massive.
I love how he owned just not knowing the ref. So rare to see. So many people would have back pedaled and been like, "oh yah, I remember now"
IMO if you have a successful product with reliable customers, a usage pattern w/o extreme peaks, and someone who knows what they're doing, on-prem makes much more financial sense. Not every company meets those criteria, and that's perfectly ok. But for those that do, building on Kubernetes opens the door of possibilities with tons of really good open source projects, and you can likely reach a density and cost advantage thats just not possible with the cloud tax. I think Kubernetes simplifies on-prem operations more than it complicates things. My current job is moving in this direction, and even at small scale the numbers just make more sense.
Meanwhile, people complaining here in the comments that Ruby is slow. While they’re reinventing the wheel for the 115th time in the newest next gen SSR JS Blazingly Fast Framework, only to notice that in real projects it is the DB that is the bottleneck.
It depends quite a bit on the application.
"So long and thanks for all the fish" comes from the insanely good Hitchhiker's Guide to the Galaxy.
All the complaints about having to replace a broken server would make sense if cloud costs were really controlable.
I’d rather roll the dice on having to buy new hardware than on AWS sending me a bill with 1000x of my typical spending.
Ah but one requires endless approvals and business justifications just to be canceled anyway. While the other just happens.
So long and thanks for all the fish is actually The Hitchhicker's Guide To The Galaxy reference. I think that was the title of the 4th book in the series. Brilliant British humour. The source of the "Answer to the great question of the universe, life and everything" bit.
Colocation is not "on-prem". Colocation means renting rackspace, power, IP transit and NOC personnel from an existing data center (similar to dedicated bare-metal servers, but you own the hardware). On-premises literally means doing everything yourself in your own building.
No cloud like external 2.5" 1TB HDD
And backups?
Throws 5 32gb flash drives on the table.
@@ogopogoman4682 🤣
@@ogopogoman4682 USB 2.0, tape drives aren't fast either
Worked at at large saas. We did a cost savings push on Amazon and after 3 projects we saved 15mil a month. We did more aggressive peering over transit gateway, swapped to arm, and moved to karpender.
I joined a company that leans HARD into AWS, not just cloud but all the AWS specific primitives, Lambda, SQS, SNS, EventBridge, etc. It seemed a little coolaidy at first but actually it's been great. Our traffic is VERY bursty, we have two massive spikes every day and then basically zero, and a fraction of that over summer. We need super high uptime and reliability. The application is very interlinked and complex.
Probably 80-90% of CRUD applications that just have large storage needs don't actually need cloud in all likelihood, but in some cases it make a ton of sense.
Definitely the infra type is dependent on the application. For HPC cloud is a total no go.
For bursty traffic, you are definitely doing it right with serverless.
Found a server rack outside in NYC, paper said on it that it works, guy bought it on eBay for $350, wrote that it can be sold $200 easily, I just took it and keep it at home, but have no time to play around it
I hate that modern software engineering culture has become about all or nothing opinions. Either cloud good + on-prem is the devil, or cloud bad + on-prem is the only way. Just solve problems with the right solution. Cloud is the right solution sometimes. On-prem is the right solution other times. Cloud empowers small and early stage teams/developers. Without cloud computing, so many businesses would be dead before they even had a chance. At the same time, it is expensive and sacrifices lots of control. Decide whats right for you based on the facts of your circumstances, don't make your decisions based off of internet discourse, and what some other guy did.
True.
I was for a tjme lead architect for infrastructure at a large online game company who operated in the cloud (not Zynga, the other big one with actual games games), and there had been a preliminary decision to leave the cloud just before my hire. I was asked to analyze and build a business case either way. Buying hardware, all the premium datacenter costs, the additional maintenance was cashflow neutral end of first year and saved $20 million total expenditures over 36 months (+-25% depending on business growth assumptions and how you applied capex and opex, which the CFO hadn’t decided a standard for yet).
The $7.5 million hardware outlay was the easiest infrastructure check I ever got signed. Only glitch was that we ordered right as an industry wide SSD generation shift in production hit, and we were six weeks late due to waiting for Intel’s first thousand units early customer access to ship (only half ours). Cost savings overall hit the numbers.
I've been saying this for years. Cloud is more expensive in the long run once you move past just having a few services. I am a big believer in trying to keep as much stuff onprem as possible to keep costs low and you don't have to spend all the time moving back and forth.
Cloud has that sweet spot where if you are too big then it's better to get your own hardware
"Cloud is le bad" - new hot take every jumps on to smug up the room
For now I'm cloud based. When your operating expenses for AWS are still under $100 a month with over $150K ARR, you're tiny. But there is no way going to self-hosted would be cost or time effective at this point. Not when you're the only person (4 hours east of ThePrimeagen) in the middle of corn fields.
With the hitchhiker guide I believe there’s a complete video of the author reading it on UA-cam. Really enjoy the authors intonation of the characters. Found it more enjoyable than the movie if I would be honest.
You have other options, you can move to a COLO... Reminder, if you on Cloud, you still have infrastructure engineers. Cloud is just expensive.. I work for a company that uses AWS, Google and Microsoft azure. you talked about S3, you can do that on prem.
I saw the same story in another company that I worked for as a contractor. Basically no cloud = huge savings. Cloud is probably only good to start or as a backup infra. If you start in cloud and grow enough, then the time will come to leave it.
18 petabytes of flash storage is crazy. DHH should Collab with LTT once everything is set up
"So long and take all the fish" is a reference to the superior intelligent beings leaving Earth (dolphins) before destruction on D. Adams book.
I specialize in Cloud cost optimization (FinOps). It’s crazy to see how many companies are not utilizing the cost optimization levers in the cloud. Often because of a lack of responsibility for the costs.
I've heard this topic 1 or 2 years ago. Oh boy this is going to pop.
We did our best to leave the MSOffice cage and people can't help but find another one. Hopeless.
I recommend listening to The Hitchhiker's Guide to the Galaxy because Douglas Adams originally wrote it as a radio show. It sounds a bit better than it reads.
Thank to the person that mentioned Hitchhiker's guide to the galaxy. It was nagging at the back of my mind
My team of 8 people including an EM manages ~ 6000 VMs and 3500 macminis hosted on company owned and third party DCs and we save a ton by not hosting on cloud. If your work load is stable it pays to be not on cloud, ops would be high but savings is still substantial
1 minute content, 12 minutes gibber gabber
One rack of servers with 20TB drives is 3.3PB of disk space, 1PB with triple mirroring. Rookie numbers.
End statement is a reference to Hitchhikers guide to the galaxy. Great book. Read it dude
Glad to see I am not the only person that took Donald Rumsfeld’s motto to heart.
Love to see it. Ready for "just someone else's server" craze/rush to crest and decline.
it does not make sense because having your own physical servers is always more expensieve, power, admin sys, more people and contractors. It is always cheap when your business is on apps and websites rather than the whole operations.
in a former company we spent more time in migrating between cloud providers and authentication services than we spent in building the products.
Government not seizing but instead heavily regulating that kind of business is a common take in most countries including the US since when a business adopts a monopolistic position on a market kinda like it does in the electric generation and distribution sector its often heavily regulated and striped from its "supper-normal profits" (referring characteristic of a monopolistic market in microeconomics) so the price is equal to the marginal cost of that good or service.
Its not communism, its pragmatism. The US, Canada, France, UK, and many more countries do that with for example electricity generation and other naturally monopolistic sectors.
On the other hand it could be feasible that the market ends up regulating itself (if that were the case we could be certain that its not a monopoly). It would be caused by people just leaving the cloud and going to on premise, or instead just by the different cloud providers competing against each other.
Managing 3 devs is not a full time job, unless you seriously hired the wrong people. Unless you’re a small startup, you probably have an existing manger or even the CTO who can handle an extra couple of meetings per week plus whatever review process and admin busywork for 3 more staff.
It would be unlikely you’d hire a new manager for such a small team. Usually they farm those duties out to a team lead which is an awful role but getting popular due to the cost cutting of making a senior dev pull triple duty as a dev, manager, and usually a PM as well.
this video is nonsense from the start. i couldn't finish it. it is not worthy of my time and life...
They usually totally forget about all of the labor & taces and ancillary costs like insurance, power, redundant fiber uplink, etc. The hardware is the cheapest part of the equation.
There are 3 types of lies - lies, damned lies, and cloud cost estimates
You've been lied to for years - especially for a small startup that might never actually need to scale.
Has not read Hitchhiker's? ... stop everything you are doing and rectify.
DHH before Neovim: 🤡
DHH after Neovim(same take): 💪
No matter what you do, just make sure you use Neovim
nonsense. he’s reasonable ever since he put the finger up the j2ee folks with rails.
I thought everyone knew that cloud doesn't save money, it saves time. My company is trying to move from On-Prem to cloud and we have spent a year spending 10x the quoted cost from AWS and the performance is still worse.
The Cloud hype is the new Microsservices hype. Hope people realize finally it's not a silver bullet. It has tradeoffs and it's not always worth it. Never underestimate private services costs and architectural complexity
14:46 probably also don't understand 42
9:55 I mean I get why they hate DHH as I did too originally but the more I saw of his takes the more of it seemed fine. There are still many takes I disagree with but most of the takes that get surfaced are ones I would disagree with in the first place. Those opinions you never hear of are the one you will agree with. I can't hate someone for having takes I disagree with, at least not when it's not the majority of the takes.
1:55 I find this to be an out dated view of the company hierarchy. You're correct that this is how it would have been done traditionally. However I would say in general if employees are professionals such as engineers then there is less of a requirement for oversight. Instead you can have a lead professional do the necessary required work to facilitate the business function all whilst carrying out the same work as the other non lead professionals. Any specific employment issues would be dealt by HR as they most likely would've been if a manager was involved.
The advantage of having a lead that also does the job is that they'll also understand the issues on the "shop floor" and may also be able to be proactive in resolving them.
AWS cloud architect, most of the time on prem is probably what you need. Large conpanies that have a lot of red tape benefit from the cloud as their change process relies more heavily on third party vendors which is actually a very large body of work.
Bursty workloads do better in the cloud but the vast majority of applications arent actually that bursty.
With K8s its much easier to lift and shift things around to other providers to get the best bang for your buck.
Last sentence refers to line of dialogue in Hitchhikers guide to the galaxy by dolphins (if I remember it correctly).
Big cloud is lying to us 🤯
If you need to dynamically scale, use cloud. Otherwise, onprem will always be cheaper.
Hosting yourself vs cloud is renting a car vs leasing.
The last sentence is from the Hitchiker's Guide To the Galaxy. You should read it. I don't want to spoil it here. :)
I bet Primeagen doesn't even have a towel! 🤣
Nuthin like a good DHH article, Hey!
With a 100% of certainty, Those not knowing linux in software world with no hardware perspective are not only Jnior devs but also weak devs and forever be if continued without learning.
Don't use IaaS unless you have to. There is a tipping point where the volume is so high that IaaS clusters could be more cost-effective than serverless solutions like Lambda, State Machines, and Fargate containers. However, most organizations will not reach that tipping point. When people leave machines running 24x7 under-utilized, you can expect a nice fat bill. Also entrust it to people who are well aware of costing strategies and which services to use.
I hope everyone doesn’t have short term memory of what AWS did to Parlor. It was likely illegal but since their politics aligned with the ruling class they got a pass. Unless forced I will NEVER use AWS. One day they could 🤬 you over too.
Thinking not using cloud services is “thinking different”.
Translation: I’ve been doing for less than 10 years and think IT didn’t exist before I learned about it.
You don't actually have to read The Hitchhiker's Guide. It's great but you can also choose the old TV show or the Radio show.
If you do stuff in the cloud, you still need knowledgeable people, Cloud is not a way to shed employees.