Some Companies Save $12M a Year with Baremetal! How Much Can Yours?
Hi. I'm Max Clark. Have you ever noticed that anyone offering an AWS cost reduction service is going to quote you something along the lines of 30 to 40%. Whatever you're spending on AWS will save you 30 to 40%. And if you see this, you probably wonder.
Speaker 1:It's like, is that real? It seems not right. Are they lying to me? What's going on here? Let me just get right to it.
Speaker 1:Yes. They're lying to you. The real numbers are so much higher that they're actually shocking. So people are afraid to use it because it seems so completely insane that nobody would believe it. Real numbers, like, maybe 80%.
Speaker 1:And actually, just to back up a second before you start going down this path, there are lots of tooling that you should look at in terms of AWS cost reduction, cost management. So are you tagging your resources? Are you aligning tag resources to what your actual business is doing? You know, can you actually do cost allocation tags and say this application generates this much revenue and costs as much to run? If you haven't done that stuff, you should start there.
Speaker 1:Start with a tagging strategy and actually equate how you're spending money on your cloud platform to what revenue is being generated off of that spend. It'll be interesting. It'll be probably illuminating to your team if you haven't done that yet. Anyways, within AWS, I mean, there's tools that'll help you find unattached EBS volumes. Guaranteed you have some.
Speaker 1:You're gonna find stuff where, like, easy 2 isn't running. Guaranteed you're gonna find some. You're gonna find instant sizing recommendations guaranteed. You're gonna find some. You're gonna have s 3 bucket issues.
Speaker 1:Basically, everything you're gonna use, there's gonna be something that should just be done differently. And that's just normal. And yes, you should figure out a way to actually run this at scale because as your cloud infrastructure increases, it becomes harder and harder to manage. And you need to use a tool to actually help you manage it. If you're not using that tool or if you're not using a tool, by the way, I'm not even talking about savings plans, reserved instances, you know, you can or or EDPs, you know?
Speaker 1:So start with a tool, start with just going through and and understanding how you're spending and if you're spending the money on something you shouldn't be spending money on and and changing how you're spending money. And then after you do that, I would tell you I would say, you know, investigate savings plans, investigate our reserve instances. Be really careful with EPs. They're the devil. And that'll help knock down what you're spending.
Speaker 1:By the way, you probably don't want to get into one of these platforms. It's going to tell you we're going to save you money by automatically turning stuff off in the middle of the night. I mean, maybe you have a workflow that's really don't need it at night and you should turn it off, but chances are it's probably not the case. Okay. You can save a ton of money there.
Speaker 1:Let's talk about how you really save money when you're looking at a AWS to bare metal migration. The first thing is it's really difficult to understand what the actual AWS instant sizing means for you. And deep in the bowels of AWS, there's a footnote that explains a vCPU or subscription on course of 3 to 1. I've been looking for this for years, and I was actually amazed to find this the other day. So let's just go with a 3 x oversubscribe between core count and vCPU.
Speaker 1:So that'll give you kind of a little bit of sizing guidance on what you would need in bare metal. Now most virtualization platforms, when you go into a virtualization platform, let's say, you know, VMware, Hyper V, whatever it is, you can probably run these things at 4 to 6 x. So you can double the vCPU count. I'm not gonna use those numbers for this. I'm just gonna tell you that's available.
Speaker 1:So the numbers can actually get even better. Let's take the biggest instance type that you can actually, let's just this. I'm gonna use a real client example, and I'm gonna explain this line by line and and kind of work through this. Because if you're looking at it solely from a, oh, we're running this little instance and it only costs us, you know, $500 a month. And if we ran this instance somewhere else, it would probably cost us.
Speaker 1:That's a fool's errand. It's a distraction prop. You're thinking about this wrong. Take a step back. Let me run you through this.
Speaker 1:Okay. So in this case, this client was a I3, I3en and I3en metal consumer. Why? Well, they were running large containers with a lot of disk and network IO, and they had a very intensive application. Now, if you go into any of the Kubernetes platforms, you start talking about container sizing and node sizing.
Speaker 1:You know, there's different strategies for how big of your instance you want, how many containers you put on your instance, where it saturates out, how much hardware real utilization you get, all these different things, how much overhead you have from the AWS Nitro platform, which, by the way, is awesome. I'm not saying it's not awesome. It is awesome what they've done. And you get to be sizing things. So I'm just gonna run you through this.
Speaker 1:AWS I three e and metal instance is 36 core or 96 v CPU, 768 gig of RAM. It has 87.5 terabyte NVMe drives and a 100 gig network interface. K. So 96 v c v CPU, 768 gig of RAM, roughly 30 terabytes raw NVMe. I remember advertised disk sizes in 1,000 versus 1024s, and you're going to put a RAID set on it.
Speaker 1:And so you're not really getting 30 terabytes, but we'll say 30 terabytes and it has a 100 gig NIC. AWS is on demand price for this is $7,924.30 a month. That's on demand. You go into 1 year. No upfront r I.
Speaker 1:You're down to $5,396.82 Right. That's a pretty decent discount. And you can get into a savings plans for 15 to 20% discount. And there's ways to get into savings plans with a 15 to 20% discount without paying an upfront charge on it. You're just committing to capacity.
Speaker 1:So same thing you're going from. You end up at basically the same place now. Yes. You can purchase 3 year non convertible paid upfront our eyes and you can get cheaper pricing, but nobody actually does that. Your finance department is not gonna let you do it.
Speaker 1:You're not gonna wanna do it. Your CTO is not gonna do it. If you're the CTO, you're not signing it. Nobody does these sorts of things. So that is a fake number.
Speaker 1:Ignore it. We use the 1 year no upfront our eye cost because it usually gives the best pricing example of what the real world is in AWS. So we're just gonna say it's $54100 a month for that instance. 96 CPU going to a bare metal instance. 64 core AMD processor at a 3x250.
Speaker 1:Actually, in this case, I ended up after testing at a 4x 256vCPU, 1 terabyte of RAM, 10 7.6 gig NVMe drives and 4 100 gig NICs And the 4 100 gig NICs were done so that way they could pair the public side and pair the private side and have redundancy on both interfaces plus additional capacity of the network because of the way the application was working. This application, after going through a proof of concept, was sized to provide roughly 3x the capacity per node for their Kubernetes environment versus what they were getting from AWS. That's how we got to a CPU configuration, core count, RAM disk space, NIC interfaces was roughly to get to a 3 x. So, you know, if you're running one instance on on the AWS infrastructure, you're running 3 instances on bare metal. Right.
Speaker 1:Just a 3 x. They have way more capacity in 3 x. We're just going to give you the numbers for 3 x. By the way, the cost for that bare metal node is $1,126 per month. AWS 1 year RI 5396 bare metal 11.26 plus a 3 x increase in density.
Speaker 1:Do you see the number? Right. Do you get it? Like, I I can probably put this we'll put this on the screen. Actually, we haven't even gotten the worst part yet.
Speaker 1:But anyways, to summarize, 3 x the container density per node, 79% less money per node. It's just bonkers. Now let's talk about egress. AWS, if you're just looking at it from a computer storage place play, there are configurations where the pricing absolutely makes sense and the cost of taking out of s 3 or out of e c 2 for your application, maybe it doesn't make sense. That's the distraction problem where AWS is crushing it or crushing you in terms of money is data.
Speaker 1:Moving data from availability zones inside of a region, NAT gateways, egress being the big one. Now AWS advertises a 5¢ per gig egress. If you're at scale, you're gonna be on an EDP. You're gonna have something else. You're gonna pay a lot less.
Speaker 1:In this case, I'm gonna use these numbers at 20 petabytes egress, 2.7¢ per gig, and that is $540,000 a month. And you're gonna say to yourself, hey, that's 2.7¢ per gig. That's that's almost half, you know, the 5¢ per gig. That's a fantastic price. Absolutely.
Speaker 1:You did a great job negotiating almost 50% reduction in your egress cost. You went from what would have been over $1,000,000 a month in egress to $540,000 a month in egress, and you should be happy and you should pat yourself in the back. And now I'm gonna tell you to put that in a data center on redundant 100 gig network interfaces. It's going to cost you $16,250 per month. 20 petabytes of egress translates to 65 gigabits on 90 5th percentile.
Speaker 1:And by the way, there are network operators out there that'll sell you this stuff at 100 gig flat rate network interfaces for $5,000 a month. This pricing decreases month over month, year over year. These are numbers today. May 1, 2024, $540,000 a month, AWS, $16,250 a month, bare metal. If you've been following along, now scale this pricing up.
Speaker 1:Under these instances in AWS plus bandwidth egress, $1,060,000 per month, AWS, bare metal, 30 instances, 30 nodes, plus the bandwidth, $50,000 a month. Do you wanna spend $12,960,000 a year on AWS, or do you wanna spend $60,000 per year on bare metal? What would you do with an extra $12,000,000 a year in your budget? Like, materialized out of thin air. $12,000,000 a year.
Speaker 1:More profit or projects, hire more team engineers, salespeople, marketing, events, activities. Now maybe you're a $50,000,000,000 company and these numbers are peanuts. They don't make any sense to you. Fine. The scale of what you're spending on AWS is just much larger.
Speaker 1:It's probably not $12,000,000. It's probably a $100,000,000 a year in that case. Again, maybe it's peanuts. Maybe it doesn't make sense for you. But if you're a startup and you're talking about what's your burn rate, what's your burn multiple.
Speaker 1:Right? Or do you run out of cash in a bank? Here's $12,000,000 a year that comes back into you that can extend your runway and do other things with it. If you are a sane business that needs to generate profit in order to stay in business, here's we're $12,000,000 a year. Yes.
Speaker 1:Come after me. Shots fired. Come after me. Tell me I'm wrong and I'm an idiot. But fine.
Speaker 1:Here's the thing. You know, if you're incubating software projects and you're going through rapid development and you're, you know, trying to figure out, you know, what platform or what infrastructure you need, absolutely 100% use a cloud. Don't believe though that the elasticity and you need to scale up and scale down, most environments don't do that. There's a relatively steady state. Okay.
Speaker 1:Fine. Figure out, you know, if you need a 100% utilization and you're afraid of doing this, very commonly, we'll start people off at, like, 50 to 60% of their total capacity. And we'll put that in bare metal and we'll have overflow in AWS. You can link the 2 environments together. You can link your bare metal with your AWS environment, and it works fine.
Speaker 1:You know? It was a good example of this. 37 signals launched in the email platform. Hey. If you haven't used this, pretty cool.
Speaker 1:But they talk a lot about how they went from AWS. They didn't know what this thing was gonna be. They didn't know what they were gonna need. They didn't know anything about it. They were writing a platform from scratch.
Speaker 1:They started in AWS. They ran. They launched. They got traffic. They figured out what their profile was.
Speaker 1:They they know what their capacity needs are. They moved into a data center. They're saving a ton of money. By the way, that money goes the owners of the business and also gets shared with the employees. Like, that money, instead of being paid into AWS, it's not being paid into their own pockets.
Speaker 1:Right? Look, if you're small and you're spending under a few $100,000 a year in AWS spend, like, maybe this doesn't make sense for you. You know, it depends on how what what kind of operations and infrastructure you told you need and what your development plans are and what you're doing. I mean, there's lots of different things. I mean, this stuff, none of these decisions are like absolute.
Speaker 1:They become really specific to the situation. Once you cross at like 50 or $100,000 a month in in cloud spend, you should start thinking about this and you should start paying really close attention to it and you should start tracking the alternatives. You should understand what savings plan your eye costs are. You should have a tool in place that gives you the ability to actually manage costs. You should be doing cost tagging and tracking actual expense to revenue within your cloud platform.
Speaker 1:And then you should also just kind of be aware of what would it cost to move it off AWS and run it somewhere else. And you should keep this going. And, you know, I've heard this express, take your total infrastructure cost and express it as a percentage of total revenue. Right? And we find companies all the time that'll come back and say, hey, you know, we're we're spending 12, 14, 15, 16 percent of our revenue on our cloud infrastructure, and it's the business economics.
Speaker 1:The fundamentals are upside down. And what do we do about it? Well, you know, if you do something like this, if it's here, all of a sudden you're talking about going from 16% of your spend to 6% of your revenue by spend, you know, in in your business, like, what does that do for your economics? Like, it radically changes how the business functions. Anyways, I will probably never stop ranting about this.
Speaker 1:So anyways, circle back around. Yes. I'll never stop ranting about this. The numbers are crazy, but 30% is just something that gets told to people because anything more than that feels insane. I will tell you from practice.
Speaker 1:In most cases, we end up somewhere in that, like, 70 to 80 percent number if we really go all the way. And by the way, bare metal, you're not buying the hardware. You don't have a capex, you know, expense on it. You're not maintaining the hardware. You don't have to put people in a data center.
Speaker 1:You're not signing a data center contract. This is still managed capacity infrastructure. If you have a big deployment and you go into a data center, a co location environment where you buy the equipment, put it on a lease, sign a contract with a data center, get transit providers, you know, hire somebody to help you manage it. If it's not close to you, the numbers go lower. You get to a point where it's like, does it even need to be lower?
Speaker 1:I mean, if you're saving $12,000,000 a year, if you could save another $1,000,000 a year, does that extra $1,000,000 will be mattered? For most people at that point, they're like, hey, you know, let's not take on the operational overhead because it's just it's okay. We did enough, you know, like that'll do pig. Right? One of those kind of things.
Speaker 1:I'm Max Clark. I ran about this all the time. If you have any questions, comment, DM, email, call me. Of course, you know, I'm expecting some hate. So, you know, leave that below as well.
Speaker 1:Hope this helps you.
