Podcasts Technology Screaming in the Cloud Cutting Costs in Cloud with Everett Berry

Cutting Costs in Cloud with Everett Berry

Published: May 9, 2023, 10 a.m.

b'

Everett Berry, Growth and Open Source at Vantage, joins Corey at Screaming in the Cloud to discuss the complex world of cloud costs. Everett describes how Vantage takes a broad approach to understanding and cutting cloud costs across a number of different providers, and reveals which providers he feels generate large costs quickly. Everett also explains some of his best practices for cutting costs on cloud providers, and explores what he feels the impact of AI will be on cloud providers. Corey and Everett also discuss the pros and cons of AWS savings plans, why AWS can\\u2019t be counted out when it comes to AI, and why there seems to be such a delay in upgrading instances despite the cost savings.


About Everett

Everett is the maintainer of ec2instances.info at Vantage. He also writes about cloud infrastructure and analyzes cloud spend. Prior to Vantage Everett was a developer advocate at Arctype, a collaborative SQL client acquired by ClickHouse. Before that, Everett was cofounder and CTO of Perceive, a computer vision company. In his spare time he enjoys playing golf, reading sci-fi, and scrolling Twitter.

Links Referenced:



Transcript

Announcer: Hello, and welcome to Screaming in the Cloud with your host, Chief Cloud Economist at The Duckbill Group, Corey Quinn. This weekly show features conversations with people doing interesting work in the world of cloud, thoughtful commentary on the state of the technical world, and ridiculous titles for which Corey refuses to apologize. This is Screaming in the Cloud.

Corey:\\xa0 LANs of the late 90\\u2019s and early 2000\\u2019s were a magical place to learn about computers, hang out with your friends, and do cool stuff like share files, run websites & game servers, and occasionally bring the whole thing down with some ill-conceived software or network configuration. That\\u2019s not how things are done anymore, but what if we could have a 90\\u2019s style LAN experience along with the best parts of the 21st century internet? (Most of which are very hard to find these days.) Tailscale thinks we can, and I\\u2019m inclined to agree. With Tailscale I can use trusted identity providers like Google, or Okta, or GitHub to authenticate users, and automatically generate & rotate keys to authenticate devices I\'ve added to my network. I can also share access to those devices with friends and teammates, or tag devices to give my team broader access. And that\\u2019s the magic of it, your data is protected by the simple yet powerful social dynamics of small groups that you trust.Try now - it\'s free forever for personal use. I\\u2019ve been using it for almost two years personally, and am moderately annoyed that they haven\\u2019t attempted to charge me for what\\u2019s become an essential-to-my-workflow service.

Corey: Have you listened to the new season of Traceroute yet? Traceroute is a tech podcast that peels back the layers of the stack to tell the real, human stories about how the inner workings of our digital world affect our lives in ways you may have never thought of before. Listen and follow Traceroute on your favorite platform, or learn more about Traceroute at origins.dev. My thanks to them for sponsoring this ridiculous podcast.\\xa0


Corey: Welcome to Screaming in the Cloud. I\\u2019m Corey Quinn. This seems like an opportune moment to take a step back and look at the overall trend in cloud\\u2014specifically AWS\\u2014spending. And who better to do that than this week, my guest is Everett Berry who is growth in open-source over at Vantage. And they\\u2019ve just released the Vantage Cloud Cost Report for Q1 of 2023. Everett, thank you for joining me.


Everett: Thanks for having me, Corey.


Corey: I enjoy playing slap and tickle with AWS bills because I am broken in exactly that kind of way where this is the thing I\\u2019m going to do with my time and energy and career. It\\u2019s rare to find people who are, I guess, similarly afflicted. So, it\\u2019s great to wind up talking to you, first off.


Everett: Yeah, great to be with you as well. Last Week in AWS and in particular, your Twitter account, are things that we follow religiously at Vantage.


Corey: Uh-oh [laugh]. So, I want to be clear because I\\u2019m sure someone\\u2019s thinking it out there, that, wait, Vantage does cloud cost optimization as a service? Isn\\u2019t that what I do? Aren\\u2019t we competitors? And the answer that I have to that is not by any definition that I\\u2019ve ever seen that was even halfway sensible.


If SaaS could do the kind of bespoke consulting engagements that I do, we would not sell bespoke consulting engagements because it\\u2019s easier to click button: receive software. And I also will point out that we tend to work once customers are at a certain point at scale that in many cases is a bit prohibitive for folks who are just now trying to understand what the heck\\u2019s going on the first time finance has some very pointed questions about the AWS bill. That\\u2019s how I see it from my perspective, anyway. Agree? Disagree?


Everett: Yeah, I agree with that. I think the product solution, the system of record that companies need when they\\u2019re dealing with Cloud costs ends up being a different service than the one that you guys provide. And I think actually the to work in concert very well, where you establish a cloud cost optimization practice, and then you keep it in place via software and via sort of the various reporting tools that the Vantage provide. So, I completely agree with you. In fact, in the hundreds of customers and deals that Vantage has worked on, I don\\u2019t think we have ever come up against Duckbill Group. So, that tells you everything you need to know in that regard.


Corey: Yeah. And what\\u2019s interesting about this is that you have a different scale of visibility into the environment. We wind up dealing with a certain profile, or a couple of profiles, in our customer base. We work with dozens of companies a year; you work with hundreds. And that\\u2019s bigger numbers, of course, but also in many cases at different segments of the industry.


I also am somewhat fond of saying that Vantage is more focused on going broad in ways where we tend to focus on going exclusively deep. We do AWS; the end. You folks do a number of different cloud providers, you do Datadog cost visibility. I\\u2019ve lost track of all the different services that you wind up tracking costs for.


Everett: Yeah, that\\u2019s right. We just launched our 11th provider, which was OpenAI and for the first time in this report, we\\u2019re actually breaking out data among the different clouds and we\\u2019re comparing services across AWS, Google, and Azure. And I think it\\u2019s a bit of a milestone for us because we started on AWS, where I think the cost problem is the most acute, if you will, and we\\u2019ve hit a point now across Azure and Google where we actually have enough data to say some interesting things about how those clouds work. But in general, we have this term, single pane of glass, which is the idea that you use 5, 6, 7 services, and you want to bundle all those costs into one report.


Corey: Yeah. And that is something that we see in many cases where customers are taking a more holistic look at things. But, on some level, when people ask me, \\u201cOh, do you focus on Google bills, too,\\u201d or Azure bills in the early days, it was, \\u201cWell, not yet. Let\\u2019s take a look.\\u201d And what I was seeing was, they\\u2019re spending, you know, millions or hundreds of millions, in some cases, on AWS, and oh, yeah, here\\u2019s, like, a $300,000 thing we\\u2019re running over on GCP is a proof-of-concept or some bizdev thing. And it\\u2019s\\u2026 yeah, why don\\u2019t we focus on the big numbers first? The true secret of cloud economics is, you know, big numbers first rather than alphabetical, but don\\u2019t tell anyone I told you that.


Everett: It\\u2019s pretty interesting you say that because, you know, in this graph where we break down costs across providers, you can really see that effect on Google and Azure. So, for example, the number three spending category on Google is BigQuery and I think many people would say BigQuery is kind of the jewel of the Google Cloud empire. Similarly for Azure, we actually found Databricks showing up as a top-ten service. Compare that to AWS where you just see a very routine, you know, compute, database, storage, monitoring, bandwidth, down the line. AWS still is the king of costs, if you will, in terms of, like, just running classic compute workloads. And the other services are a little bit more bespoke, which has been something interesting to see play out in our data.


Corey: One thing that I\\u2019ve heard that\\u2019s fascinating to me is that I\\u2019ve now heard from multiple Fortune 500 companies where the Datadog bill is now a board-level concern, given the size and scale of it. And for fun, once I modeled out all the instance-based pricing models that they have for the suite of services they offer, and at the time was three or $400 a month, per instance to run everything that they\\u2019ve got, which, you know, when you look at the instances that I have, costing, you know, 15, 20 bucks a month, in some cases, hmm, seems a little out of whack. And I can absolutely see that turning into an unbounded growth problem in kind of the same way. I just\\u2026 I don\\u2019t need to conquer the world. I\\u2019m not VC-backed. I am perfectly content at the scale that I\\u2019m at\\u2014


Everett: [laugh].


Corey: \\u2014with the focus on the problems that I\\u2019m focused on.


Everett: Yeah, Datadog has been fascinating. It\\u2019s been one of our fastest-growing providers of sort of the \\u2018others\\u2019 category that we\\u2019ve launched. And I think the thing with Datadog that is interesting is you have this phrase cloud costs are all about cloud architecture and I think that\\u2019s more true on Datadog than a lot of other services because if you have a model where you have, you know, thousands of hosts, and then you add-on one of Datadogs 20 services, which charges per host, suddenly your cloud bill has grown exponentially compared to probably the thing that you were after. And a similar thing happens\\u2014actually, my favorite Datadog cost recommendation is, when you have multiple endpoints, and you have sort of multiple query parameters for those endpoints, you end up in this cardinality situation where suddenly Datadog is tracking, again, like, exponentially increasing number of data points, which it\\u2019s then charging to you on a usage-based model. And so, Datadog is great partners with AWS and I think it\\u2019s no surprise because the two of them actually sort of go hand-in-hand in terms of the way that they\\u2026 I don\\u2019t want to say take ad\\u2014


Corey: Extract revenue?


Everett: Yeah, extract revenue. That\\u2019s a good term. And, you know, you might say a similar thing about Snowflake, possibly, and the way that they do things. Like oh, the, you know, warehouse has to be on for one minute, minimum, no matter how long the query runs, and various architectural decisions that these folks make that if you were building a cost-optimized version of the service, you would probably go in the other direction.


Corey: One thing that I\\u2019m also seeing, too, is that I can look at the AWS bill\\u2014and just billing data alone\\u2014and then say, \\u201cOkay, you\\u2019re using Datadog, aren\\u2019t you?\\u201d Like, \\u201cHow did you know that?\\u201d Like, well, first, most people are secondly, CloudWatch is your number two largest service spend right now. And it\\u2019s the downstream effect of hammering all the endpoints with all of the systems. And is that data you\\u2019re actually using? Probably not, in some cases. It\\u2019s, everyone turns on all the Datadog integrations the first time and then goes back and resets and never does it again.


Everett: Yeah, I think we have this set of advice that we give Datadog folks and a lot of it is just, like, turn down the ingestion volume on your logs. Most likely, logs from 30 days ago that are correlated with some new services that you spun up\\u2014like you just talked about\\u2014are potentially not relevant anymore, for the kind of day-to-day cadence that you want to get into with your cloud spending. So yeah, I mean, I imagine when you\\u2019re talking to customers, they\\u2019re bringing up sort of like this interesting distinction where you may end up in a meeting room with the actual engineering team looking at the actual YAML configuration of the Datadog script, just to get a sense of like, well, what are the buttons I can press here? And so, that\\u2019s\\u2026 yeah, I mean, that\\u2019s one reason cloud costs are a pretty interesting world is, on the surface level, you may end up buying some RIs or savings plans, but then when you really get into saving money, you end up actually changing the knobs on the services that you\\u2019re talking about.


Corey: That\\u2019s always a fun thing when we talk to people in our sales process. It\\u2019s been sord\\u2014\\u201cAre you just going to come in and tell us to buy savings plans or reserved instances?\\u201d Because the answer to that used to be, \\u201cNo, that\\u2019s ridiculous. That\\u2019s not what we do.\\u201d But then we get into environments and find they haven\\u2019t bought any of those things in 18 months.


Everett: [laugh].


Corey: \\u2014and it\\u2019s well\\u2026 okay, that\\u2019s step two. Step one is what are you using you shouldn\\u2019t be? Like, basically measure first then cut as opposed to going the other direction and then having to back your way into stuff. Doesn\\u2019t go well.


Everett: Yeah. One of the things that you were discussing last year that I thought was pretty interesting was the gp3 volumes that are now available for RDS and how those volumes, while they offer a nice discount and a nice bump in price-to-performance on EC2, actually don\\u2019t offer any of that on RDS except for specific workloads. And so, I think that\\u2019s the kind of thing where, as you\\u2019re working with folks, as Vantage is working with people, the discussion ends up in these sort of nuanced niche areas, and that\\u2019s why I think, like, these reports, hopefully, are helping people get a sense of, like, well, what\\u2019s normal in my architecture or where am I sort of out of bounds? Oh, the fact that I\\u2019m spending most of my bill on NAT gateways and bandwidth egress? Well, that\\u2019s not normal. That would be something that would be not typical of what your normal AWS user is doing.


Corey: Right. There\\u2019s always a question of, \\u201cAm I normal?\\u201d is one of the first things people love to ask. And it comes in different forms. But it\\u2019s benchmarking. It\\u2019s, okay, how much should it cost us to service a thousand monthly active users? It\\u2019s like, there\\u2019s no good way to say that across the board for everyone.


Everett: Yeah. I like the model of getting into the actual unit costs. I have this sort of vision in my head of, you know, if I\\u2019m Uber and I\\u2019m reporting metrics to the public stock market, I\\u2019m actually reporting a cost to serve a rider, a cost to deliver an Uber Eats meal, in terms of my cloud spend. And that sort of data is just ridiculously hard to get to today. I think it\\u2019s what we\\u2019re working towards with Vantage and I think it\\u2019s something that with these Cloud Cost Reports, we\\u2019re hoping to get into over time, where we\\u2019re actually helping companies think about well, okay, within my cloud spend, it\\u2019s not just what I\\u2019m spending on these different services, there\\u2019s also an idea of how much of my cost to deliver my service should be realized by my cloud spending.


Corey: And then people have the uncomfortable realization that wait, my bill is less a function of number of customers I have but more the number of engineers I\\u2019ve hired. What\\u2019s going on with that?


Everett: [laugh]. Yeah, it is interesting to me just how many people end up being involved in this problem at the company. But to your earlier point, the cloud spending discussion has really ramped up over the past year. And I think, hopefully, we are going to be able to converge on a place where we are realizing the promise of the cloud, if you will, which is that it\\u2019s actually cheaper. And I think what these reports show so far is, like, we\\u2019ve still got a long ways to go for that.


Corey: One thing that I think is opportune about the timing of this recording is that as of last week, Amazon wound up announcing their earnings. And Andy Jassy has started getting on the earnings calls, which is how you know it\\u2019s bad because the CEO of Amazon never deigned to show up on those things before. And he said that a lot of AWS employees are focused and spending their time on helping customers lower their AWS bills. And I\\u2019m listening to this going, \\u201cOh, they must be talking to different customers than the ones that I\\u2019m talking to.\\u201d Are you seeing a lot of Amazonian involvement in reducing AWS bills? Because I\\u2019m not and I\\u2019m wondering where these people are hiding.


Everett: So, we do see one thing, which is reps pushing savings plans on customers, which in general, is great. It\\u2019s kind of good for everybody, it locks people into longer-term spend on Amazon, it gets them a lower rate, savings plans have some interesting functionality where they can be automatically applied to the area where they offer the most discount. And so, those things are all positive. I will say with Vantage, we\'re a cloud cost optimization company, of course, and so when folks talk to us, they often already have talked to their AWS rep. And the classic scenario is, that the rep passes over a large spreadsheet of options and ways to reduce costs, but for the company, that spreadsheet may end up being quite a ways away from the point where they actually realize cost savings.


And ultimately, the people that are working on cloud cost optimization for Amazon are account reps who are comped by how much cloud spending their accounts are using on Amazon. And so, at the end of the day, some of the, I would say, most hard-hitting optimizations that you work on that we work on, end up hitting areas where they do actually reduce the bill which ends up being not in the account manager\\u2019s favor. And so, it\\u2019s a real chicken-and-egg game, except for savings plans is one area where I think everybody can kind of work together.


Corey: I have found that\\u2026 in fairness, there is some defense for Amazon in this but their cost-cutting approach has been rightsizing instances, buy some savings plans, and we are completely out of ideas. Wait, can you switch to Graviton and/or move to serverless? And I used to make fun of them for this but honestly that is some of the only advice that works across the board, irrespective in most cases, of what a customer is doing. Everything else is nuanced and it depends.


That\\u2019s why in some cases, I find that I\\u2019m advising customers to spend more money on certain things. Like, the reason that I don\\u2019t charge percentage of savings in part is because otherwise I\\u2019m incentivized to say things like, \\u201cBackups? What are you, some kind of coward? Get rid of them.\\u201d And that doesn\\u2019t seem like it\\u2019s going to be in the customer\\u2019s interest every time. And as soon as you start down that path, it starts getting a little weird.


But people have asked me, what if my customers reach out to their account teams instead of talking to us? And it\\u2019s, we do bespoke consulting engagements; I do not believe that we have ever had a client who did not first reach out to their account team. If the account teams were capable of doing this at the level that worked for customers, I would have to be doing something else with my business. It is not something that we are seeing hit customers in a way that is effective, and certainly not at scale. You said\\u2014as you were right on this\\u2014that there\\u2019s an element here of account managers doing this stuff, there\\u2019s an [unintelligible 00:15:54] incentive issue in part, but it\\u2019s also, quality is extraordinarily uneven when it comes to these things because it is its own niche and a lot of people focus in different areas in different ways.


Everett: Yeah. And to the areas that you brought up in terms of general advice that\\u2019s given, we actually have some data on this in this report. In particular Graviton, this is something we\\u2019ve been tracking the whole time we\\u2019ve been doing these reports, which is the past three quarters and we actually are seeing Graviton adoption start to increase more rapidly than it was before. And so, for this last quarter Q1, we\\u2019re seeing 5% of our costs that we\\u2019re measuring on EC2 coming from Graviton, which is up from, I want to say 2% the previous quarter, and, like, less than 1% the quarter before. The previous quarter, we also reported that Lambda costs are now majority on ARM among the Vantage customer base.


And that one makes some sense to me just because in most cases with Lambda, it\\u2019s a flip of a switch. And then to your archival point on backups, this is something that we report in this one is that intelligent tiering, which we saw, like, really make an impact for folks towards the end of last year, the numbers for that were flat quarter over quarter. And so, what I mean by that is, we reported that I think, like, two-thirds of our S3 costs are still in the standard storage tier, which is the most expensive tier. And folks have enabled S3 intelligent tiering, which moves your data to progressively cheaper tiers, but we haven\\u2019t seen that increase this quarter. So, it\\u2019s the same number as it was last quarter.


And I think speaks to what you\\u2019re talking about with a ceiling on some cost optimization techniques, where it\\u2019s like, you\\u2019re not just going to get rid of all your backups; you\\u2019re not just going to get rid of your, you know, Amazon WorkSpaces archived desktop snapshots that you need for some HIPAA compliance reason. Those things have an upper limit and so that\\u2019s where, when the AWS rep comes in, it\\u2019s like, as they go through the list of top spending categories, the recommendations they can give start to provide diminishing returns.


Corey: I also think this is sort of a law of large numbers issue. When you start seeing a drop off in the growth rate of large cloud providers, like, there\\u2019s a problem, in that there are only so many exabyte scale workloads that can be moved inside of a given quarter into the cloud. You\\u2019re not going to see the same unbounded infinite growth that you would expect mathematically. And people lose their minds when they start to see those things pointed out, but the blame that oh, that\\u2019s caused by cost optimization efforts, with respect, bullshit it is. I have seen customers devote significant efforts to reducing their AWS bills and it takes massive amounts of work and even then they don\\u2019t always succeed in getting there.


It gets better, but they still wind up a year later, having spent more on a month-by-month basis than they did when they started. Sure they understand it better and it\\u2019s organic growth that\\u2019s driving it and they\\u2019ve solved the low hanging fruit problem, but there is a challenge in acting as a boundary for what is, in effect, an unbounded growth problem.


Everett: Yeah. And speaking to growth, I thought Microsoft had the most interesting take on where things could happen next quarter, and that, of course, is AI. And so, they attributed, I think it was, 1% of their guidance regarding 26 or 27% growth for Q2 Cloud revenue and it attributed 1% of that to AI. And I think Amazon is really trying to be in the room for those discussions when a large enterprise is talking about AI workloads because it\\u2019s one of the few remaining cloud workloads that if it\\u2019s not in the cloud already, is generating potentially massive amounts of growth for these guys.


And so, I\\u2019m not really sure if I believe the 1% number. I think Microsoft may be having some fun with the fact that, of course, OpenAI is paying them for acting as a cloud provider for ChatGPT and further API, but I do think that AWS, although they were maybe a little slow to the game, they did, to their credit, launch a number of AI services that I\\u2019m excited to see if that contributes to the cost that we\\u2019re measuring next quarter. We did measure, for the first time, a sudden increase on those new [Inf1 00:20:17] EC2 instances, which are optimized for machine learning. And I think if AWS can have success moving customers to those the way they have with Graviton, then that\\u2019s going to be a very healthy area of growth for them.


Corey: I\\u2019ll also say that it\\u2019s pretty clear to me that Amazon does not know what it\\u2019s doing in its world of machine-learning-powered services. I use Azure for the [unintelligible 00:20:44] clients I built originally for Twitter, then for Mastodon\\u2014I\\u2019m sure Bluesky is coming\\u2014but the problem that I\\u2019m seeing there is across the board, start to finish, that there is no cohesive story from the AWS side of here\\u2019s a picture tell me what\\u2019s in it and if it\\u2019s words, describe it to me. That\\u2019s a single API call when we go to Azure. And the more that Amazon talks about something, I find, the less effective they\\u2019re being in that space. And they will not stop talking about machine learning. Yes, they have instances that are powered by GPUs; that\\u2019s awesome. But they\\u2019re an infrastructure provider and moving up the stack is not in their DNA. But that\\u2019s where all the interest and excitement and discussion is going to be increasingly in the AI space. Good luck.


Everett: I think it might be something similar to what you\\u2019ve talked about before with all the options to run containers on AWS. I think they today have a bit of a grab bag of services and they may actually be looking forward to the fact that they\\u2019re these truly foundational models which let you do a number of tasks, and so they may not need to rely so much on you know, Amazon Polly and Amazon Rekognition and sort of these task-specific services, which to date, I\\u2019m not really sure of the takeoff rates on those. We have this cloud costs leaderboard and I don\\u2019t think you would find them in the top 50 of AWS services. But we\\u2019ll see what happens with that.


AWS I think, ends up being surprisingly good at sticking with it. I think our view is that they probably have the most customer spend on Kubernetes of any major cloud, even though you might say Google at first had the lead on Kubernetes and maybe should have done more with GKE. But to date, I would kind of agree with your take on AI services and I think Azure is\\u2026 it\\u2019s Azure\\u2019s to lose for the moment.


Corey: I would agree. I think the future of the cloud is largely Azure\\u2019s to lose and it has been for a while, just because they get user experience, they get how to talk to enterprises. I just\\u2026 I wish they would get security a little bit more effectively, and if failing that, communicating with their customers about security more effectively. But it\\u2019s hard for a leopard to change its spots. Microsoft though has demonstrated an ability to change their nature multiple times, in ways that I would have bet were impossible. So, I just want to see them do it again. It\\u2019s about time.


Everett: Yeah, it\\u2019s been interesting building on Azure for the past year or so. I wrote a post recently about, kind of, accessing billing data across the different providers and it\\u2019s interesting in that every cloud provider is unique in the way that it simply provides an external endpoint for downloading your billing data, but Azure is probably one of the easiest integrations; it\\u2019s just a REST API. However, behind that REST API are, like, years and years of different ways to pay Microsoft: are you on a pay-as-you-go plan, are you on an Azure enterprise plan? So, there\\u2019s all this sort of organizational complexity hidden behind Azure and I think sometimes it rears its ugly head in a way that stringing together services on Amazon may not, even if that\\u2019s still a bear in and of itself, if you will.


Corey: Any other surprises that you found in the Cloud Cost Report? I mean, looking through it, it seems directionally aligned with what I see in my environments with customers. Like for example, you\\u2019re not going to see Kubernetes showing up as a line item on any of these things just because\\u2014


Everett: Yeah.


Corey: That is indistinguishable from a billing perspective when we\\u2019re looking at EC2 spend versus control plane spend. I don\\u2019t tend to [find 00:24:04] too much that\\u2019s shocking me. My numbers are of course, different percentage-wise, but surprise, surprise, different companies doing different things doing different percentages, I\\u2019m sure only AWS knows for sure.


Everett: Yeah, I think the biggest surprise was just the\\u2014and, this could very well just be kind of measurement method, but I really expected to see AI services driving more costs, whether it was GPU instances, or AI-specific services\\u2014which we actually didn\\u2019t report on at all, just because they weren\\u2019t material\\u2014or just any indication that AI was a real driver of cloud spending. But I think what you see instead is sort of the same old folks at the top, and if you look at the breakdown of services across providers, that\\u2019s, you know, compute, database, storage, bandwidth, monitoring. And if you look at our percentage of AI costs as a percentage of EC2 costs, it\\u2019s relatively flat, quarter over quarter. So, I would have thought that would have shown up in some way in our data and we really didn\\u2019t see it.


Corey: It feels like there\\u2019s a law of large numbers things. Everyone\\u2019s talking about it. It\\u2019s very hype right now\\u2014


Everett: Yeah.


Corey: But it\\u2019s also\\u2014you talk to these companies, like, \\u201cOkay, we have four exabytes of data that we\\u2019re storing and we have a couple 100,000 instances at any given point in time, so yeah, we\\u2019re going to start spending $100,000 a month on our AI adventures and experiments.\\u201d It\\u2019s like, that\\u2019s just noise and froth in the bill, comparatively.


Everett: Exactly, yeah. And so, that\\u2019s why I think Microsoft\\u2019s thought about AI driving a lot of growth in the coming quarters is, we\\u2019ll see how that plays out, basically. The one other thing I would point to is\\u2014and this is probably not surprising, maybe, for you having been in the infrastructure world and seeing a lot of this, but for me, just seeing the length of time it takes companies to upgrade their instance cycles. We\\u2019re clocking in at almost three years since the C6 series instances have been released and for just now seeing C6 and R6 start to edge above 10% of our compute usage. I actually wonder if that\\u2019s just the stranglehold that Intel has on cloud computing workloads because it was only last year around re:Invent that the C6in and the Intel version of the C6 series instances had been released. So, I do think in general, there\\u2019s supposed to be a price-to-performance benefit of upgrading your instances, and so sometimes it surprises me to see how long it takes companies to get around to doing that.


Corey: Generation 6 to 7 is also 6% more expensive in my sampling.


Everett: Right. That\\u2019s right. I think Amazon has some work to do to actually make that price-to-performance argument, sort of the way that we were discussing with gp2 versus gp3 volumes. But yeah, I mean, other than that, I think, in general, my view is that we\\u2019re past the worst of it, if you will, for cloud spending. Q4 was sort of a real letdown, I think, in terms of the data we had and the earnings that these cloud providers had and I think Q1 is actually everyone looking forward to perhaps what we call out at the beginning of the report, which is a return to normal spend patterns across the cloud.


Corey: I think that it\\u2019s going to be an interesting case. One thing that I\\u2019m seeing that might very well explain some of the reluctance to upgrade EC2 instances has been that a lot of those EC2 instances are databases. And once those things are up and running and working, people are hesitant to do too much with them. One of the [unintelligible 00:27:29] roads that I\\u2019ve seen of their savings plan approach is that you can migrate EC2 spend to Fargate to Lambda\\u2014and that\\u2019s great\\u2014but not RDS. You\\u2019re effectively leaving a giant pile of money on the table if you\\u2019ve made a three-year purchase commitment on these things. So, all right, we\\u2019re not going to be in any rush to migrate to those things, which I think is AWS getting in its own way.


Everett: That\\u2019s exactly right. When we encounter customers that have a large amount of database spend, the most cost-effective option is almost always basically bare-metal EC2 even with the overhead of managing the backup-restore scalability of those things. So, in some ways, that\\u2019s a good thing because it means that you can then take advantage of the, kind of, heavy committed use options on EC2, but of course, in other ways, it\\u2019s a bit of a letdown because, in the ideal case, RDS would scale with the level of workloads and the economics would make more sense, but it seems that is really not the case.


Corey: I really want to thank you for taking the time to come on the show and talk to me. I\\u2019ll include a link in the [show notes 00:28:37] to the Cost Report. One thing I appreciate is the fact that it doesn\\u2019t have one of those gates in front of it of, your email address, and what country you\\u2019re in, and how can our salespeople best bother you. It\\u2019s just, here\\u2019s a link to the PDF. The end. So, thanks for that; it\\u2019s appreciated. Where else can people go to find you?


Everett: So, I\\u2019m on Twitter talking about cloud infrastructure and AI. I\\u2019m at@retttx, that\\u2019s R-E-T-T-T-X. And then of course, Vantage also did quick hot-takes on this report with a series of graphs and explainers in a Twitter thread and that\\u2019s @JoinVantage.


Corey: And we will, of course, put links to that in the [show notes 00:29:15]. Thank you so much for your time. I appreciate it.


Everett: Thanks, Corey. Great to chat.


Corey: Everett Berry, growth in open-source at Vantage. I\\u2019m Cloud Economist Corey Quinn and this is Screaming in the Cloud. If you\\u2019ve enjoyed this podcast, please leave a five-star review on your podcast platform of choice, whereas if you\\u2019ve hated this podcast, please leave a five-star review on your podcast platform of choice along with an angry, insulting comment that will increase its vitriol generation over generation, by approximately 6%.


Corey: If your AWS bill keeps rising and your blood pressure is doing the same, then you need The Duckbill Group. We help companies fix their AWS bill by making it smaller and less horrifying. The Duckbill Group works for you, not AWS. We tailor recommendations to your business and we get to the point. Visit duckbillgroup.com to get started.

'