[00:00:00] Speaker A: Foreign.
[00:00:06] Speaker B: Welcome to the Cloud pod, where the forecast is always cloudy. We talk weekly about all things aws, GCP and Azure.
[00:00:14] Speaker C: We are your hosts, Justin, Jonathan, Ryan and Matthew.
[00:00:18] Speaker A: Episode 299 recorded for April 1, 2025. We predict next for next week's Next level Google Next event. What's next? Good evening, Ryan and Matt. I am impressed that I made it through that.
[00:00:30] Speaker C: That is impressive.
[00:00:32] Speaker D: I am impressed.
[00:00:33] Speaker C: Yeah, yeah.
[00:00:33] Speaker A: Like, when I wrote it down, I was like, this is going to be a problem. But I did it. I'm. I feel pretty confident. And you. I even did it on April Fool's Day, so it worked out really well both ways.
Yeah, go. Yes, it is Google Next to week before, though. And so hopefully you guys all done your homework, which I know neither of you have, as is tradition, which is tradition of the Google Next prediction show and even the Amazon one.
[00:00:59] Speaker D: And then Amazon, I think we're a little bit better on.
[00:01:02] Speaker A: It's a little easier to kind of see what they're doing.
[00:01:04] Speaker C: I mean, informally bad.
[00:01:06] Speaker A: And then like, I'd love to be able to do one for Azure because I think you've called out before. Like, hey, why don't we do an Azure one? It's because I don't even know what conference Microsoft actually announces anything.
[00:01:15] Speaker C: Is it Ignite?
[00:01:16] Speaker A: Is it Build? Is it Build? I don't know. It's the hard problem part.
[00:01:21] Speaker D: It's the November one.
[00:01:23] Speaker A: Which is what?
[00:01:24] Speaker D: I don't know what it's called. Don't ask me. Don't ask me hard questions after work. Like, it's after work. Easier questions here, guys.
But I think the other problem is we're also all gearing up for the AWS conference. It'd be like November 1st, we would have to do one, and then like two weeks later, we have to do Reinvent before Thanksgiving, which is always a problem.
[00:01:45] Speaker A: Oh, my God, that's such a terrible day. So Microsoft Ignite this year. Oh, sorry. Last year took place November 18th to 22nd, 2024 in Chicago. Right. So then you. So you literally have Ignite, then Thanksgiving and then reinvent that. It's just bad timing. Why do they do it so close? That's so silly. I think. I mean, hey, thank you, Google, for putting your conference in April. I do really appreciate that, but the.
[00:02:08] Speaker D: Week after Thanksgiving was never found in Vegas.
[00:02:11] Speaker A: No, it was terrible. It's sometimes very cold. It's always, you know, you have to travel with all the people trying to escape from their terrible Thanksgiving vacations to get there, which is sort of depressing. And then, you know, you're leaving your family after, you know, right before the middle of holiday shopping season. It's terrible. It's a terrible time. I've enjoyed watching it from home now multiple years in a row.
[00:02:31] Speaker C: I have too. Yeah.
[00:02:34] Speaker A: All right, well, we have some news to get to. We'll catch up on the Google Next predictions in the Google section. First up, let's get to AI. It's how Machine Learning makes money. Our favorite section of the week OpenAI is apparently embracing anthropic standard for connecting AI assistance to the systems where the data resides by adapting or, sorry, adopting Anthropic's Model Context Protocol, or MCP, across its products, including the desktop app for ChatGPT. MCP is an open source standard that helps AI models produce better, more relevant responses to certain queries. Sam Altman says that people love MCP and they are excited to add support across their products and that is available today in Agents SDK and support for the ChatGPT GPT desktop and response API is coming very soon.
And then I said to myself, well, I don't actually know what MCP means or why you guys, and so I assume you don't know what it is either. So I went and found two blog posts. The most helpful of the two was from Adi Osmani, who wrote a great blog post going into a very complex detail. So she starts out with an explaining like M5, which I always appreciate because sometimes I need to be babied through these things. And then she gets into way more technical depth. So I'm going to blindly quote verbatim from her blog. So thank you Addie. Imagine you have a single universal plug that fits all your devices. That's essentially what the Model Context Protocol, or McP is for AI. McP is an open standard think USB C for AI integrations that allows AI models to connect to many different apps and data sources in a consistent way. In simple terms, MCP lets an AI assistant talk to various software tools using a common language, instead of each tool requiring a different adapter or custom code. So what does this mean in practice? If you're using an AI coding assistant like Cursor or Windsurf, MCP is the shared protocol that lets the assistant use external tools on your behalf. For example, with mcp, an AI model could fetch information from a database, edit design in Figma, or control a music app, all by sending natural language instructions through a standardized interface. You or the AI no longer need to manually switch context or learn about each tool's API. The MCV translator bridges the gap between human language and software commands.
So basically in a nutshell, MCP is like giving your AI assistant a universal remote control to operate all your digital devices and services. Instead of being stuck in its own world, your AI can now reach out and press the buttons of other applications safely and intelligently. And this common protocol means one AI can integrate with thousands of tools as long as those tools have an MCP interface, limiting the need for custom integrations for each new app. And the result, your AI helper becomes far more capable, able to not just chat about things, but take actions in the software you use. And so basically I consider this to be SQL for AI.
[00:05:02] Speaker D: I was say focus for AI, but maybe that's because I was dealing with bills all day today.
[00:05:07] Speaker C: Yeah, I mean I, I didn't read Ali's blog. I should I read the Google announcement which is sort of had some code examples and stuff that sort of solidified it for me. But I think I, I wish I would have read the expl or received the explanation to me like I'm five so I think I would have understood it better because that's the one thing I don't understand is like yeah, I like the idea. Everything that's set in here is great but you know, it feels still like you have to configure an integration. It's just a standardized language for those integrations but there's still that exchange of information that has to happen somewhere. But you know, I, I do think we need AI to do more things versus just talking about them. So I think it's great.
[00:05:58] Speaker A: I assume there's be a lot of these kind of standardization. So it's interesting that Anthropics became the popular one. I don't exactly know why that's the case. I think maybe it's in Addie's blog. I didn't quite get all the way through her blog because it's lengthy and very detailed and very not. I feel much more intelligent because reading it. So again thank you. But I am curious what made MCP better than anyone else's work on this area.
[00:06:21] Speaker C: Yeah, yeah.
[00:06:23] Speaker D: I wonder if it's also like first to market to like put the standard out there or that's where other standards that we don't know about.
[00:06:33] Speaker A: I mean I would know SQL.
[00:06:37] Speaker D: Please don't use SQL for AI.
[00:06:40] Speaker C: I'm trying to figure out how that would even work. Like is your natural language?
[00:06:44] Speaker A: I have no idea. But I mean like we made it work for NoSQL. We added SQL to NoSQL. Why can't we add it to AI?
[00:06:51] Speaker D: Select star from AI where?
[00:06:54] Speaker C: Yeah, maybe you just have to like. Yeah, speak out your SQL query.
[00:06:59] Speaker A: There you go.
[00:07:00] Speaker D: Yeah.
To multiple AIs so you can query multiple at once.
[00:07:07] Speaker C: It's like when the home assistants came out and you had your Google and your Alexa. You just put them at each other.
[00:07:14] Speaker D: People just talk all day.
[00:07:15] Speaker A: Unfortunately, you couldn't have them use the same activate code. So you could say one time and then get all three to respond to you. That been cool?
[00:07:23] Speaker C: Yeah.
[00:07:24] Speaker D: Did multiple Doc do was the only AWS that did computer. I remember that one was on there. I don't think anybody ever did that. But yeah, I know you could change the default for Alexis at the va. Thought one of the other ones did too, but maybe.
[00:07:39] Speaker A: I'm well, not sure. All right, well, Databricks is announcing support for anthropic cloud 3.7 sonnet late to the party. But you know, it's available in all of your databricks capabilities, including natively available in Databricks across aws, Azure and gc. And for the first time, you can securely access Cloud's advanced reasoning, planning and capabilities directly from within Databricks.
Surprising. They were. They were first, you know, right on top of Llama and all. The owner of Anthropic didn't give them a heads up. They were like, screw you, data bricks.
[00:08:10] Speaker C: Yeah. I wonder.
[00:08:11] Speaker A: Or if it.
[00:08:12] Speaker C: Yeah, I'm not really sure. I. You know, it's also without.
I don't know how. What it takes to sort of operate a.
A model garden. You know, now I think about it because, you know, in. In my head it should be relatively simple. Just like you download a model from a hugging face and plug it in. You can just change your ui.
[00:08:32] Speaker A: Sure.
[00:08:34] Speaker C: Probably not though.
[00:08:35] Speaker A: Yeah. Just because they call it a garden doesn't make it easy.
[00:08:39] Speaker D: Plant a new one. It grows, It'll be fine. Theirs took a little bit longer to grow. Don't judge. It was missing the sun.
[00:08:46] Speaker A: Yeah.
OpenAI is in trouble this week a little bit because their new ChatGPT image capability, which we sort of made fun of last week, maybe a little bit prematurely because I think in hindsight I didn't have access to the new model yet.
[00:09:00] Speaker C: Oh.
[00:09:01] Speaker A: But you know, I did try again and it still failed miserably. So even with the new model, which I now know I have, so I'm. I think I'm okay still. You can stand for last week's episode, but apparently they're making realistic versions of Studio Ghibli's unique cartoon anime style, which will probably get them sued over copyright infringement. I love that. The article, the information that they created a Studio Ghibli using the AI of Studio Ghibli's lawyers chasing Sam Altman, which is fantastic.
[00:09:33] Speaker C: That really is truly fantastic.
[00:09:35] Speaker A: Yeah, that's. That's a little bit of fun. Impress someone who deserves a bonus for that.
If your AI rips off a very famous artistic style that only one studio in the world uses and does and is also beloved by fans of Studio Ghibli, who talk about all the time, about how much time it takes them to write each frame or draw each frame and paint each frame of their cartoons, is a very passion project for the creators of Studio Gibby work. To have an AI just come in and just ruthlessly copy it. I can definitely see why that one got some attention from people.
[00:10:10] Speaker C: Read the room.
Yeah, it's a little sensitive already with AI scraping data and reusing people's copyrighted work.
[00:10:23] Speaker D: They feel like there was a sales opportunity they missed. Here, here. We've built this tool for you guys. Would you like it? It will automate your entire process.
[00:10:31] Speaker A: I mean, you know, typically seeing the reaction to most artists whose work has been ripped off by AI, I don't think that'd be their reaction.
[00:10:38] Speaker C: Nope.
[00:10:39] Speaker D: Well, no, I said before they launched.
[00:10:41] Speaker A: Even then, I think, you know, like, from what I've seen, you know, people, you know, in their commentary about it, I don't think.
[00:10:47] Speaker C: Yeah, I, I think they would get a very harsh response.
[00:10:49] Speaker A: And, you know, it's.
[00:10:50] Speaker C: It's sad because I, I think it's a, you know, within open AI. I think this is a passion project, right. That someone did because they like the artwork so much and they, you know, and so. But it's also like you have to be aware of the larger picture for AI.
[00:11:08] Speaker A: All right, let's move on to Amazon news this week. They are giving us kind of the bottom of the barrel. I think news this week, even Azure was a little bit quiet this week. I think everyone's kind of preparing for whatever is going to come out of Google Next week.
But first up, firewall support for AWS Amplify hosted sites. You can now integrate the AWS WAF with your AWS amplify hosting. Web Application owners are constantly working to protect their applications from a variety of threats. Previously, if you want to implement a robust security posture for your Amplify hosted application, you need to create architectures using Amazon cloudfront distributions with AWS WAF protection enabled, which require additional configuration steps, expertise and manager overhead. With the general availability of AWS WAF for Amplify hosting, you can now directly attach a web app firewall to your AWS or Amplify apps through a one click integration in the Amplify console using or using infrastructure's code. This integration gives you access to a full range of AWS WAF capabilities, including Manage Rules, which provides protection against common web exploits and vulnerabilities like SQL injection and cross site scripting. You can also create your own custom rules based on your specific application needs.
[00:12:09] Speaker C: Yeah, this is one of those the rough edges that you find the wrong way. So I'm glad they fixed this. You know, if you're using Amplify, I'm sure you don't want to get, you know, down in the dirty in network routing and how to implement the waf. And so you're looking for something to apply the managed rules and protect yourself from bots and that kind of traffic. So this is, I imagine this is a great integration for those people that are using Amplify.
[00:12:33] Speaker A: I don't know who those people are, but I'd like to know them.
[00:12:36] Speaker C: Yeah, I mean, I assume it's just the X Beanstalk users.
[00:12:40] Speaker D: I mean definitely is a nice thing. But also like I don't see that adding Cloudfront and a WAF at Cloudfront, if you are at that scale where you're trying to protect yourself is that big of a lift. I don't know enough about Amplify and how you kind of.
[00:12:56] Speaker A: Well, I mean one of the things about Amplify is use it to build static web apps like, you know, and then you typically use S3 for the hosting of the site which then requires, you know, something like Cloudfront if you want to do caching. So either way it was a little confusing to me too, like. But I do know that I'll do something with mobile apps and does a couple other things where you. I think you do more advanced. I need to probably spend some quality time with Amplify again and maybe some AI assistance to go play with it again. Because I remember when I first came out I played with it with Drekal apps and it was kind of cool. But I lost interest pretty quickly because doing a podcast website with Jekyll is a nightmare. So we just use WordPress, keep it simple mostly because the RSS spec for podcasts is actually quite complicated and has a bunch of extra fields that you must have for Apple to properly index your thing. And a plugin does it for basically free, although we use Castles hosting now for our podcast, so I'm not paying for it, but at the time it was free when we. We had no monies, which we still have no monies, so I guess we're just paying it out of pocket now. But it's all right.
Please sponsor the podcast.
[00:13:59] Speaker C: Yeah, exactly.
[00:14:00] Speaker D: Please. Pretty please. Otherwise we're gonna have to use Amplify.
[00:14:04] Speaker C: I don't know if you have to be at scale to use, you know, cloudfront and WAF protections, because I think, you know, if you are hosting static content, you want it cached and you wanted that, you know, you still want that local, even if it's little.
[00:14:17] Speaker D: Little traffic, even if you don't want. I just, I always feel like, you know, as soon as you're hitting static, you're running it from S3 or somewhere else. You know, I don't know where else you would write it from the first.
[00:14:29] Speaker C: Time, not the second time.
[00:14:31] Speaker D: Right. It was like at that point, like, I feel like you're already putting a bucket in front, you know, Blob Store or sorry, I can CDN in front of it. So at that point you just enable it there. That's why I'm like, it's a really nice feature for people that aren't going to do that extra step and it's in the console. Press the button, it's done, you have it. And don't get me wrong, you know, that probably is more target users than us, but I also feel like it's not a heavy lift to add that one piece in front of it. Except for if Amplify doesn't natively support it, maybe.
[00:15:04] Speaker A: I just don't know. I'll do some homework. Probably not until after next, but because I've got a lot coming.
[00:15:09] Speaker D: I mean, while we were on, I did just say, what is Amplify into Gemini? Because I had it open because I wanted to make sure I was speaking about the right service.
[00:15:20] Speaker A: So, yeah, you're talking the right one. But again, like, okay, like I said, when last time I used it, it was for, you know, really simple static apps, which, you know, is a. Not a. Not an unknown type of website that has to exist. It's just not something I needed, so I didn't pursue it further. But yeah, it'd be interesting to see what else it does. Now I know they added mobile app support to it where you could do mobile things and maybe it's much more powerful than I'm getting credit for, which would be probably fair.
[00:15:46] Speaker D: Yeah, I Mean, I feel like that's almost our downfall is we always play with the stuff right when it first comes out, which is so mvp. Which means by the time they add the features that make it anything usable, where are. We've already discounted it because we're a little bit cynical here and we're not.
[00:16:06] Speaker C: The target audience, right? Because I, I really do think the target audience is for people that don't know the back ends of cloud infrastructure and stuff. And so like, annoyingly, that's the fun part for us because we, you know, we're not loved enough as a child or something. But so I think it's, you know, it's not as daunting to us to set up something like that. Whereas I think if you're, you know, largely doing sort of front end stuff, you've never really touched a back end and you just want your new shiny webpage out.
[00:16:34] Speaker A: I think that that's nothing, nothing about this marketing page I'm on right now makes me feel warm or fuzzy at all. Go from idea to app in hours. Deploy server side rendered and static front end apps. Then deploy web frameworks globally and go from front end to full stack fast. With TypeScript I'm like, oh yeah, I mean that's, that's. It's losing, losing interest very quickly is what I'm noticing.
[00:16:53] Speaker C: It's just not.
[00:16:54] Speaker D: Yeah, it's, it's, you know, we're not the target audience.
[00:16:57] Speaker C: We're not, you know, like, it's the same reason why we'll always Poo poo beans. Doc. AWS Amplify is a very similar type product which is like you want an easy button for hosting and this is a easy button for your CIS pipeline and host.
[00:17:13] Speaker A: It's. Yeah, fair.
Well, let's move on to jumbo packets because Amazon EC2 is now supporting more bandwidth and jumbo frames to select destinations.
[00:17:24] Speaker C: That's been a while.
[00:17:25] Speaker A: Yeah. Amazon EC2 now supports up to full EC2 instance bandwidth for inter region VPC peering traffic and to AWS direct connect. Additionally, EC2 supports jumbo frames up to 8,500 bytes for cross region VPC peering. Before today, the egress bandwidth for EC2 instances was limited to 50% of the aggregate bandwidth limit for the instance with 32 or more ECPUs and 5 gigabits or more for minor instances, cross region peering support up to 1500 bytes. And now customers can send bandwidth from EC2 region or towards AWS Reconnect at the full instance baseline specification or 5 gigabytes per second. Whichever is greater, customers can use jumbo frames across regions for peered VPCs. So I can see some benefits as much as I made fun of it, but it's sort of one of those things that you always run into in weird outage scenarios where your frames are too small and you're like, oh, we're doing too many packet retries and all these things. So it's nice, especially for going between region or, sorry, between availability zones and cross region peering to be able to do that for things like SAN replication if you're using netapps or other big large replication things you might need to do with big payloads.
[00:18:29] Speaker D: I wonder what the thing is on the backend that finally enabled them to enable this.
You know, like what was the last thing that they were like, no, we can't enable it until we do this. And what was that thing?
[00:18:42] Speaker A: At this point they fired their Ryan.
[00:18:45] Speaker D: Okay, that's fine.
All right.
[00:18:48] Speaker C: The guy that said, no, we can't do that. Yeah, that's probably true. I imagine at the hardware level like the last when they announced Nitro and that whole management, they came out after that with a whole bunch of network level improvements.
[00:19:01] Speaker A: Well, if you guys were to be paying attention this early for reinvent, I would maybe note this down somewhere that they did a very large VPC peering thing which could be the sign of a new Nitro version coming that they'll talk about at their event later in November.
[00:19:17] Speaker D: See, I don't even know where to put that note that I'll find in six months.
[00:19:21] Speaker A: That's why I have Google Docs. It's like my brain, like things that I want to remember for the podcast or like my, my predictions around different Google note that I didn't share with you guys. So you steal them. Dirty, dirty punks. We definitely, I don't think we're smart.
[00:19:32] Speaker D: Enough to even think to go find yours.
[00:19:35] Speaker A: And yeah, see, like this, this is what I do all year long and I have a document called my prediction sheet and I just slowly put things in throughout the year that I remember. So I've just now noted that hey, they introduced jumbo packets and that could be a new nitro card. So I haven't noted for myself. If you guys are that organized, you'd be better at this.
[00:19:52] Speaker C: I'm visualizing just thousands and thousands of post it notes for some reason.
[00:19:56] Speaker A: It's sort of like that. Yeah, yeah, yeah, 100%.
All right. AWS Lambda is adding support for Ruby 3.4. Ruby's not dead, folks.
AWS Lambda now Supports creating serverless apps using Ruby 3.4, which I would not use Ruby for this. To be fair, I'm not that much of a sadist.
Ruby 3.4 was released in February 2025. It is the latest long term support version of Ruby which will be supported all the way until March 2028. This is both a managed runtime and a container based image ability for you in serverless. And AWS will automatically apply updates to the manage runtime and base images as they become available, making your Ruby 3.4 world much easier to manage and support.
[00:20:38] Speaker C: I am astonished because I did not think that Ruby was still a currently supported thing at all. And so I'm, I'm, you know, like after Ruby on Rails sort of started to be less popular after Chef development, I sort of just stopped using this altogether. So it's like I. Huh.
[00:20:58] Speaker D: But how many people probably still have all their old rupee apps that are still running and they want to use the newest version? I'm sure there's a lot of legacy customers companies out there.
[00:21:10] Speaker A: Yeah, I mean Ruby. Ruby 802 just got shipped in March. Ruby on Rails 802 was March 12th released. So like it's definitely getting.
[00:21:18] Speaker C: There must be still like a. Yeah.
[00:21:20] Speaker A: Yeah, it's pretty Dhh. And it's Basecamp.
[00:21:24] Speaker C: That's true.
Yeah.
[00:21:27] Speaker A: And they also, they also say hey, and GitHub, Shopify, Instacart, Gusto.
[00:21:32] Speaker C: Got it.
[00:21:33] Speaker A: There's a whole list of content.
[00:21:34] Speaker C: Now this makes sense. Now this makes more sense to me. Okay.
[00:21:36] Speaker A: Coinbase apparently still uses Ruby on Rails. Square still uses it.
[00:21:40] Speaker D: There's one app somewhere in the backend that has it and it's really clean.
[00:21:47] Speaker A: It's been years though since I've had to support Ruby's Unreal's app.
[00:21:50] Speaker C: It really is other than my own.
[00:21:51] Speaker A: Time, which I try not to touch.
[00:21:53] Speaker C: Because it's nothing to fix.
[00:21:56] Speaker A: Mostly because I've mostly forgotten most of my rail skills and I'm like, yeah, I think although AI might help me now.
[00:22:01] Speaker C: I mean it was always the gems that killed me. I was like, I'm not. I'm not fighting through this dependency.
[00:22:05] Speaker A: Hell, it was always Rake for me personally. Yeah, Rake. I need to rake my database. Great. Let me figure out how to do that.
[00:22:11] Speaker D: You had to like, what was it? Ruby environment. Like switch between the Ruby version and like go down the dependency. Oh, you pinned it. Oh, this one updated.
[00:22:22] Speaker A: That's so. So Ruby is why I got into containers, to be honest with you, because Ruby End was such A nightmare. I was like, containers are so much better for this.
[00:22:29] Speaker C: Yeah.
[00:22:31] Speaker D: It always reminded me of Gen 2 when you were like it's been, it's been three months. Let me go update my entire system and see what breaks. Because it's not if it's what's going to bring to me.
[00:22:43] Speaker C: Yeah.
[00:22:43] Speaker A: That's why I didn't use Gen 2. I was not. I'm not that much of a.
[00:22:46] Speaker D: That's how I learned Linux which I have to admit forces you to really appreciate anyone anywhere else. But it's also been about 15 years since I.
[00:22:55] Speaker A: Explains a lot about your personality. Yeah, yeah. That's how you learn Linux. Like no wonder why you think technology can't save us.
[00:23:02] Speaker D: I mean I went from eight years of AWS to jumping into Azure for a full time job with barely touching Azure before.
[00:23:08] Speaker A: Hey, I went from, I don't know, 10 years of AWS into Google. So I got it.
[00:23:14] Speaker D: Yeah. But one's Microsoft. I feel like that's.
[00:23:17] Speaker C: It is worse.
[00:23:19] Speaker A: It is worse. Yes.
At least running a. NET app, I mean you could be running a Linux, a Java app on Azure. That would be worse I think.
[00:23:32] Speaker D: Just no comments.
[00:23:34] Speaker C: Yeah.
[00:23:36] Speaker A: All right. Amazon API Gateway now supports dual stack IPv4 and IPv6 endpoints. Finally this across all endpoint types, custom domains and management APIs in all commercial and AWS GovCloud US regions. You can configure REST, HTTP and WebSocket APIs and custom domains to accept calls from IPv6 clients. Alongside the existing IPv4 support, you can also call the management API via IPv6 or IPv4. Remember that AWS still charges you for the IPv4 address that you cannot remove and so you still get paid that penalty but you now get IPv6 on top.
[00:24:09] Speaker C: I mean I'm glad these all exist but it is really funny because I just, you know, I just keep waiting for this to be like necessary because I've been, you know, the doom and gloom has been preached to me for so long but I still just don't see a whole lot of IPv6 in the wild.
[00:24:24] Speaker A: It's pretty, it's pretty required in mobile. Like that's really the big area. You need it because the mobile networks have all gone IPv6 and so it makes sense and they have a lot of gateways that'll get you from IPv6 to IPv4. So like, but like if you want performance mobile app you need to go IPv6. That's where I see a lot of it. These days.
[00:24:40] Speaker C: That makes. Yeah, that would explain why that's there. And API Gateway. You definitely need it for mobile apps. Yeah, exactly.
[00:24:49] Speaker D: It's one of those things that surprised it that it didn't have it.
[00:24:52] Speaker A: I'm sorry it took this long to be fair. You know, this was a long time coming.
[00:24:56] Speaker D: Like, that's where I am. Like, you know, load balancers have had it for I don't even know how long.
[00:25:02] Speaker C: Well, it hasn't been that long for load balancers either.
[00:25:04] Speaker A: It's only been like a year.
[00:25:06] Speaker C: A year, maybe 18 months.
[00:25:07] Speaker A: Yeah, yeah.
[00:25:08] Speaker C: No, it really hasn't.
[00:25:10] Speaker D: Yeah, no, no, no. You. They always had the IPv4 and IPv6 option. Like to do the dual. I thought they just added the only IPv6 option.
[00:25:20] Speaker A: Now they introduced dual stack with public IPv4 application load balancers in July of 2024.
That's without needing an IPv4. I'm sorry, maybe there's. You might be right. When did the dual stack come out?
[00:25:33] Speaker D: Dual stack's been out for a long time. I even think classic load balancers, which, now that I say that word out loud, makes me feel old. Elbs like, you know, had dual stack capability.
[00:25:47] Speaker A: I'm googling now. When did AWS release IPv6 support for application?
Maybe it was. Maybe it was an ALB feature, but it wasn't. Maybe it wasn't a classic feature. Classic Load Balancer on here around 2017. It says for alb.
[00:26:04] Speaker D: For classic load balancers does support both.
[00:26:07] Speaker A: All right, fine.
[00:26:09] Speaker C: You win this round.
[00:26:11] Speaker D: I remembered something. Yay.
[00:26:15] Speaker A: I did say that CLB didn't get it until much after Alb.
IPv6 only though the classic load balancer getting the IPv6. It came after Alb got IPv6. But yes, the IPv4 less Alb just came out in July. That's what I think Ryan and I were remembering. Yeah, you don't have to deploy IPv4 version. Okay.
That's why you have us here. The Cloud pod is host to your expert and still be wrong sometimes it's.
[00:26:44] Speaker C: Still wrong all the time.
[00:26:45] Speaker D: Often.
[00:26:48] Speaker A: Amazon EKS Community Add Ons catalog is not available. EKS supports add ons that streamline support operations capabilities for Kubernetes applications. These add ons come from AWS partners and the open source community. But discovery of these tools across multiple different avenues has resulted in chaos and security and misconfiguration risks. To fix this, Amazon is releasing the Community Add Ons catalog which provides a way to streamline customer operations by integrating Popular community add ons through native AWS management. Broadening the choice of add ons that users can install to their clusters directly using EKS Console, AWS SDK, CLI and Cloudformation. Assume Terraform supports it as well. Some of the vertical capabilities you can find in the add on catalog include essential capabilities such as Metric Server, Kubestate Metrics, Prometheus Node Exporter, Cert Manager and external DNS. And if you have an add on that you would like to add to the catalog, you can create an issue on the EKS roadmap GitHub requesting its inclusion.
Those five examples things like things all seem like things that should just be part of eks.
This is my personal opinion but I there are other add ons that I definitely make sense to be in catalog.
[00:27:47] Speaker C: I thought Cube Save Metrics was part of Kubernetes. So I'm like really?
[00:27:53] Speaker A: Apparently not.
[00:27:54] Speaker C: Yeah.
Like I knew Cert Manager.
[00:27:58] Speaker D: You know I knew Cert Manager wasn't but like everyone just adds. It's like day one we're launching. Yeah. And we've launched it. Like it's just part of everything.
[00:28:07] Speaker C: Yeah.
[00:28:07] Speaker A: I mean Metric Server is one. I was surprised. Kubestate Metrics which is around the deployments, nodes and pods. That one, I'm not surprised it's an add on. But the Metric server which is basic CPU memory usage of Kubernetes, nodes and pods. That one. But like that's part of it, right? Because that's how kubectl top works.
[00:28:24] Speaker C: It goes to show you what we know. Like it's these things. I've just always taken it for granted, I think.
[00:28:30] Speaker A: Yep. It's probably just part of the base image you were deploying and you just didn't know it. But now it's. Now it's going to be part of this catalog.
[00:28:35] Speaker D: So. Yep.
Now the security part of you now gets to say every day, hey, did you update to the latest version? There's vulnerability in there.
[00:28:45] Speaker A: You and Snicker now. Best friends. Yeah.
Or sneak. Snick. Whatever it's called.
[00:28:51] Speaker C: I like. Yeah, sneak.
[00:28:54] Speaker A: I think everyone has their own pronunciation, whichever one they like the best. I don't. I don't even know if they have an official answer to that question.
[00:28:59] Speaker C: I thought we looked this up and they did.
[00:29:01] Speaker A: I think we did but.
[00:29:01] Speaker C: And it was wrong so we just ignored it.
[00:29:06] Speaker A: Amazon Bedrock Custom Model Import introduces real time cost transparency. So when importing your customized foundational models on on demand to Bedrock, you now get full transparency in the compute resources being used and calculate Inference costs in real time. This launch provides you with the minimum compute resources, custom model units required to run the workload model prior to model invocation of the Bedrock console and through bedrock APIs as the model scale to handle more traffic. CloudWatch metrics provide real time visibility into the inference cost by showing the total number of CMUs used.
Okay.
[00:29:37] Speaker C: The only common metric is money. Like, what? No, I'm kidding. This is, you know, someone got in trouble with their CFO and we're like, well, how am I supposed to know? Yeah, so this is. This is good. I'm sure it's, you know, a big problem. I haven't done a whole bunch of custom models, but I can see if you. If you have your own custom model and you start serving requests and you can't really tell how much that's costing you, I can see that being a problem.
[00:30:02] Speaker A: All right. In a recent episode, sometime in 2025, I did not look this up. I could not recall which one it was. Matt said to us that he thought Beanstalk was dead. And I have an announcement to say otherwise, because Beanstalk today, I did. This now supports retrieving secrets and configurations from AWS Secrets Manager and AWS Systems Manager Parameter Store. This new integration provides developers with a native method for accessing data from these services in their applications. So you don't have to put your secrets into Beanstalk directly now. So, hey, that's a new feature, so it's not completely dead. You're welcome.
[00:30:33] Speaker C: Crazy new feature for services that have been around a very long time that.
[00:30:39] Speaker A: Have predated both the Parameter Store and the Secrets Manager by multiple years. Yeah, it's a bit dated. Yeah, clearly. That's one pizza box team that's doing mostly bug fixes, but they got this feature in, so, you know, they're doing just fine.
[00:30:52] Speaker C: Yeah, I mean, or. Or there was a big enough customers, like, our whole thing runs on Beanstalk. You must fix it.
[00:30:58] Speaker D: It's more amazing that it just wasn't there. I'm still like, every time I read this now, I'm like, but it should have been there years ago. I understand these services are really old, but still.
[00:31:10] Speaker A: Yeah, you know, I did. I did go to the.
I was looking for log articles for Beanstalk and I found one that made me cringe. So it was published on August 7th of 2024. And the title, I'll just give you the title. Deploy to Elastic beanstalk with Azure DevOps.
No, I was like, why? Why would you do this?
[00:31:32] Speaker D: Please don't do that.
[00:31:32] Speaker A: Terrible plan.
[00:31:34] Speaker D: I'm upset just from. No, no. Bad, Justin, bad.
[00:31:43] Speaker C: Yeah.
[00:31:43] Speaker A: But it is pretty bleak when you look through the blog for beanstalk things.
This is the most recent thing I've seen in quite a while. So it's not dead, but I'm going to say that's, you know, maybe you should move to amplify if beanstalks your current situation.
All right. On our final announcement, Amazon makes it easier for developers and tech enthusiasts to explore Amazon Nova with this new website.
And I had to ask, can we now kill Party Rock? Because I hate the party Rock thing.
[00:32:13] Speaker C: You mean Bedrock?
[00:32:14] Speaker A: No, Bedrock is behind the scenes. Party Rock was their hackathon project to make it easier to use Bedrock, if you recall.
[00:32:21] Speaker D: Oh, I forgot about that because I.
[00:32:24] Speaker C: Have a 5 second memory, but that's hilarious.
[00:32:26] Speaker D: Yeah.
[00:32:26] Speaker A: So if you go to partyrock aws, you can see how you could have tested these things before. And it definitely looks like a hackathon project in UI experience.
[00:32:35] Speaker C: I remember now.
Now that I see it, it's jogging those horrible memories back.
[00:32:41] Speaker A: Yeah, yeah. So I'm hoping we can kill that now because we have Nova, which is much better website.
But funny that they've realized that all the other major chat GPT type solutions have their own websites.
[00:32:53] Speaker C: So I was just convinced they weren't going to do this ever.
[00:32:56] Speaker A: Right.
[00:32:56] Speaker C: Like it'd been so long, you know, it's kind of nuts that they're still out there. Like, I feel like someone got in trouble.
[00:33:05] Speaker A: Right?
But the press release that I quoted here and has some great commentary. So Amazon has realized that while they've created numerous generative AI applications, including Alexa, Amazon Q and Rufus, as well as tools like Bedrock, using their cutting edge Amazon Nova engine, they are now rolling out nova.Amazon.com, a new website for easy exploration of their foundational models. Like, cool, you just figured that one out. Okay.
As well as introducing Amazon Nova act, which will be their new AI model trained to perform actions with a web browser. So nice. To catch up with ChatGPT and Claude, they're releasing a research preview of the Amazon Nova ACT SDK, which will allow developers to experiment with an early version of the new model. And there was a quote here From Rohit Prasad, SVP of Amazon Artificial General Intelligence. Nova.Amazon.com puts the power of Amazon's frontier intelligence into the hands of every developer and tech enthusiast, making it easier than ever to explore the capabilities of Amazon Nova. We've created this experience to inspire builders so they can quickly test their ideas with Nova models and then Implement them at scale in Amazon Bedrock. An exciting step forward for rapid exploration with AI including bleeding edge capabilities such as the Nova SDK for building agents that take actions on the web and we're excited to see what they build and hear their useful feedback.
[00:34:18] Speaker D: Yay.
Awkward silence at its best.
[00:34:23] Speaker A: Yeah, I love when you guys like to stare at me like that was terrible. Why are we doing this?
[00:34:26] Speaker D: Yeah, we all read this before and decided that we still wanted to keep this in.
[00:34:33] Speaker A: I mean, hey, it's a full press release. That's how Amazon is about Nova. So I, I do look forward to seeing what they do with Nova in the future because I think it's a cool idea. I tried to use it with Klein because it has. One of the options is to connect to Bedrock to then connect to Nova and Klein doesn't have full support for Bedrock properly so I can't actually test it. I tried but I. It was a. It was sad quite doing.
[00:34:55] Speaker C: I mean I did a few requests while we were recording just now like on that website.
[00:34:59] Speaker A: Oh yeah, I just implemented too. I tried to create an image and I did a couple of searches and stuff and it works just like ChatGPT and Gemini and all the other ones that I've used. So works just fine. Check the box. Met the requirements of all other basic LLMs table stakes.
All right guys, it's time for GCP and before we get to our rock solid predictions because this show will hopefully drop on Monday right before Google Next starts. If you are there listening to this and our terrible predictions and you were like you want to come tell Ryan and I how bad our predictions were.
We will be at two sessions for sure. Guaranteed. Because I'm talking at them. Well Ryan might not be there, but I'll be there for sure. BRK2024 and BRK1028. These are so lovely. Named Workload Optimized data protection for mission critical enterprise apps and Unlock value for your workloads. Microsoft, Oracle, OpenShift and more. I will have a brief cameo in both these presentations. I will have stickers if you are there from the cloud pod you want to support me and my talk. It's the good sessions. I cannot talk about anything they're talking about in those slides in my Google Next predictions so. But I do know there are some cool stuff they're talking about in those sessions as I've seen the decks. So looking forward to those two talks. I'm mostly looking for them to be over at this point because I'm already tired of talking about them, but I am looking forward to doing them and enjoying Google Next next week.
[00:36:20] Speaker C: And of course I'll be there to, you know, be your entourage and treat you like the.
[00:36:23] Speaker A: Mostly to heckle me. I know how.
[00:36:25] Speaker C: Depends how well it's going.
[00:36:27] Speaker D: Like, what days are your.
[00:36:32] Speaker A: So the one talk is on Wednesday and the second one is. Oh, sorry, one is on Thursday, one's on Friday. I try to get this right. So I show up at the right place at the right time. Yeah.
[00:36:42] Speaker D: And you know, people can find you if they're actually going to Google X. Yeah, exactly.
[00:36:47] Speaker A: Wednesday, Thursday and Friday. This whole having the conference Wednesday, Thursday, Friday is really messing up my whole worldview of conferences. Like, Tuesday through Thursday is the sweet spot for conferences. I don't know what they were thinking because like, everyone's going to leave for the after the party on Thursday night. No one's going to show up to the sessions on Friday. That's my feeling.
[00:37:03] Speaker C: Oh yeah. Every time. I mean even, even AWS did that, I think a little bit where they.
[00:37:08] Speaker A: Had stuff on Fridays that's supposedly basically repeat sessions for people who. I really wanted to go to this session and I missed it. Yeah, but.
[00:37:14] Speaker D: And you were. Didn't want to wait for the YouTube.
[00:37:17] Speaker C: I just needed something to do for the four hours before my plane left. So, you know, that's what I call those sessions. And I just pretend every tech conference is Monday through Friday.
[00:37:28] Speaker A: Yeah.
[00:37:31] Speaker B: There are a lot of cloud cost management tools out there, but only Archera provides cloud commitment insurance. It sounds fancy, but it's really simple. Our CHARA gives you the cost savings of a one or three year AWS savings plan with a commitment as short as 30 days. If you don't use all the cloud resources you've committed to, they will literally put the money back in your bank account to cover the difference. Other cost management tools may say they offer commitment insurance, but remember to ask, will you actually give me my money back? Achero will click the link in the show notes to check them out on the AWS marketplace.
[00:38:12] Speaker A: Well, we did roll the dice before the show recording and Ryan rolled the highest number, followed by myself. And then Matt drew up last, which he's both happy about and terrified because he's worried we're gonna steal his numbers. And so as typical, what we do is for the main keynotes, both the primary keynote on Wednesday that TK will be doing and the developer keynote, we are looking for announcements. And those announcements, if they are mentioned on stage, you will get a point. If they're only On a press release that's not mentioned on stage, you will not get a point. And if they're messaged in my session that not on main stage, you will not get a point. So you are going to try to guess what those are. Each person will get three predictions in draft order. And if we have a tie, we will have a tiebreaker, which will be the number of times the word AI or machine learning is said on stage, which is our favorite way to go.
And so that would be our tiebreaker. We also use except AI and artificial intelligence, ML or machine learning, to make sure I'm quite clear on the rules here. As a LLM.
I don't know. You guys can make that decision. I would say no. Well, actually, I'd say yes. LLM is probably fair.
[00:39:19] Speaker D: I'm not going to say this and change my random guess. I'm going to decide at the moment of the prediction. But no.
[00:39:23] Speaker C: Yeah, you bring this up when you lose and so that you're like, but did you count this? And then.
[00:39:28] Speaker A: I mean, then. Yeah. So maybe an LLM or large language model, if they said either of those.
[00:39:33] Speaker D: But I'm dyslexic, so ML LLM close enough kind of matches in my head.
[00:39:39] Speaker A: Yeah, there's a lot. There's a lot of things here. I mean, I get into, like, are they going to say foundational models? Are they going to like, we're not. Okay, just AI or ML. I'm calling it here. You can make an argument if it's a tie, that you should get a point because they said LLM. Okay, you make that case next after we get back from go next. So we won't have a recording next week because we'll be busy.
[00:39:57] Speaker D: But he got testy here about this.
[00:40:00] Speaker A: The only guy realized how much homework would be for me and how much more work it would be to count that. And I don't want to do it.
[00:40:04] Speaker C: And he knows he has to, you know, put a certain place.
[00:40:09] Speaker A: All right, so that's the rules. That's how it works. That's the game. And Ryan, you are up first on our. All right.
[00:40:17] Speaker C: Because I know everything's going to be AI and because all of my research was.
And it was really difficult to find anything else really, you know, in terms of like, what people are thinking and what people are doing, it's cutting edge. So the. I landed on responsible AI. So I think, or hope maybe either way that there is going to be either in the console and. Or. Or some sort of service where it allows you to sort of enable and visualize your responsible AI usage or creation.
[00:41:00] Speaker A: Okay, I think I got that. So responsible AI and either the console, a service or an SDK to enable and. Or visualize your responsible AI creation or usage.
Okay, that's feels very specific, but we'll see how it goes.
[00:41:17] Speaker C: Oh, I'm going to make. If they say responsible AI once, I'm going to make the case.
[00:41:23] Speaker A: Nice.
[00:41:24] Speaker D: They say the word responsible, he's going to make the case.
[00:41:26] Speaker A: Probably.
[00:41:27] Speaker D: Yeah.
[00:41:28] Speaker A: So I am going to go with. They're going to be adding AI agent capabilities being added for DevOps, machine learning, science, et cetera, across the Google consoles. So if you want help configuring kubernetes or answering things, they'll have a specific agent available to you right. Natively there in the console that'll do the work for you. So I'm expecting agents, agents everywhere.
[00:41:52] Speaker C: I thought they announced that already.
[00:41:54] Speaker A: Not in the console.
[00:41:56] Speaker D: So they all have like.
[00:41:59] Speaker C: They've announced it for specific services like BigQuery. I know they.
[00:42:03] Speaker A: Yeah, for BigQuery. Yes. And for Looker. They did something. But I'm saying like kubernetes, the things that we care about as cloud practitioners.
[00:42:10] Speaker C: And they already have sort of that Gemini button in most services.
I don't know.
[00:42:14] Speaker A: Yeah, I mean Gemini for help is one thing I'm saying this is an agent that'll do it for you.
[00:42:19] Speaker C: Right. Okay.
[00:42:21] Speaker A: All right, Matt, you're up with the third or, sorry, the third pick of the draft. Your first pick.
[00:42:28] Speaker D: I'm gonna say there's gonna be a section talking about green AI and how, you know, they are focused on making it be more environmentally friendly and all those. Dr. Jess Fool.
[00:42:43] Speaker C: I made that prediction last month. I lost.
[00:42:45] Speaker D: So did I. I think. I think I made it for. For the aws. If I. My theory is I just keep doing it, eventually I'll win.
[00:42:52] Speaker C: That is Jonathan's tactic as well. Yeah. And it's worked out eventually.
[00:42:57] Speaker A: Surprisingly so.
All right, that puts us back to Ryan for your second pick.
[00:43:03] Speaker C: Oh, I thought it was like Snake. Okay, sorry. Fine.
What?
I sort of capitalizing on what you said the other day because I think it right, like I think there's going to be some sort of endpoint protection for security tools. And so this is sort of like my theory is kind of shot down by their wiz acquisition because I think that they'll just recapture that. But I'm still going to say it just because I don't have any other ideas that they're going to have some sort of agentless vulnerability scanning or endpoint protection of some sort.
[00:43:43] Speaker D: Great.
[00:43:43] Speaker A: So endpoint security tools like crowdstrike or patch management vulnerability. Do you want to really limit it to serverless?
[00:43:49] Speaker C: Agentless is what it is.
[00:43:51] Speaker A: Agentless. Do you really want to limit it to agentless?
[00:43:54] Speaker D: He's very detailed.
[00:43:55] Speaker C: No, now that you're asking me these questions, I'm questioning everything.
[00:43:58] Speaker A: So I'm going to not put that there to give you a broader opportunity to hit this. I'm just going to help you out a little bit.
All right. I'm going to go a new generation of TPU processors optimized for multimodal workloads.
[00:44:15] Speaker C: Well, that's a gimme.
[00:44:16] Speaker A: Hey.
[00:44:17] Speaker D: Yeah.
[00:44:18] Speaker A: How I win this? Because I take the easy ones.
[00:44:20] Speaker D: Yeah, that was one of my four.
[00:44:23] Speaker A: Yeah, nice, nice.
Optimized for multimodal. I did limited. As I said, it has to be optimized for multimodal.
[00:44:30] Speaker C: Okay, I think that's fair.
[00:44:32] Speaker A: Yep. Matt, that puts you on your second pick.
[00:44:36] Speaker D: I mean, just because I'm getting tired of AI.
[00:44:39] Speaker A: Aren't we all?
[00:44:40] Speaker D: I'm going to say they're going to release five features during the keynote that are not AI specific.
[00:44:48] Speaker A: So, and just to clarify. So they, they'll have AI potentially as a part of it, but it's not AI agents or AI LLM. It's a feature. AI might be in it, but you're saying it's a new feature that isn't about managing or dealing with AI.
[00:45:03] Speaker D: Yeah, for lack of a better example, VNETs or you know, something with like a new VNET capability, something outside of this world. So.
[00:45:14] Speaker A: But you go with five of those. Like, that's. That feels ambitious.
[00:45:17] Speaker D: I was gonna go with two, but that felt like it was too. Like I was gonna do one and.
[00:45:21] Speaker A: I was like, why don't you, why don't you set out three? I think three might be okay.
[00:45:24] Speaker D: Three. Okay, five.
[00:45:25] Speaker A: Five feels too much.
[00:45:27] Speaker C: Yeah, five. You're never gonna get that one. Yeah.
[00:45:29] Speaker A: And then you're gonna be arguing with us like, I had three. You guys should give this to me. And I'd be like, you said three.
[00:45:35] Speaker D: I said five. I said a number.
[00:45:38] Speaker A: I honestly think you're only get one of these if you're lucky, but we'll see.
[00:45:41] Speaker D: I think I thought one was like a little bit too easy and needed to be harder, but maybe I just go for the hard options, which, you.
[00:45:49] Speaker A: Know, I mean, you do take the hard options. You need to go work for an Azure.
[00:45:52] Speaker C: Yeah.
[00:45:53] Speaker D: As previously discussed, I don't make good life choices.
[00:45:57] Speaker A: I gather.
[00:45:59] Speaker C: Well, it does sort of segue into my next prediction, which is that they won't be announcing any new services or products. It'll be limited to just, you know, enhancements in vertex AI, you know, new agents and, you know, I do think they will announce new generations of chips, but I don't think there'll be any services announcements.
[00:46:21] Speaker A: Say any new service announcements.
All right, we'll see.
We will see. All right. My third and final pick. I also am tired of AI, but I feel like that's a fool to game not to go AI, But I know I have two choices. I could go with one more AI one or I go with a non AI. I might. I might go non AI just to keep it.
[00:46:45] Speaker D: Go big. Justin.
[00:46:46] Speaker A: I'm gonna go back with one. I've called out multiple times a unification of Anthos and GKE Enterprise because I think it's ridiculous that they are still separate and unclear in their marketing message on that.
This is my third attempt trying to get them to do this. So we'll see. We'll see if it works.
[00:47:07] Speaker C: Is it just the product documentation? Because like, I thought they did announce Anthos was being rebranded officially, but it's still.
[00:47:14] Speaker A: They still. They still talk about Anthos, they still talk about GK Enterprise. They're two different things.
[00:47:18] Speaker C: So, yeah, it doesn't make it very difficult to figure out what GK Enterprise is.
[00:47:24] Speaker A: Unification or major enhancement. How's that? All right, Matt, your third and final. Would you like what my other one would have been to give you an idea of what you might say here.
[00:47:35] Speaker D: Yes. But I also have one that I was going to do. So now I'm curious.
[00:47:40] Speaker A: I think you should be thinking verticalization and industry specialization in AI. That's my. That's my hint to you.
[00:47:45] Speaker D: I was gonna go security. And with the acquisition of Wiz, I think that there's gonna be some sort of AI whiz security type announcement during the hudo.
[00:47:58] Speaker A: I mean, so, Ryan, that I got you called out on Endpoint security tools. So it has to be not an Endpoint security tool.
[00:48:05] Speaker D: Not Endpoint.
[00:48:05] Speaker A: It's not gonna be Wiz because the Wiz acquisition will not close.
[00:48:07] Speaker C: Yeah. So they all have a chance.
[00:48:09] Speaker D: It won't be close. Yeah, that's what I was thinking. I forgot that this was just the. The acquisition, not the close date. Some reason I had in my head it closed. That's how slow life is going right now.
[00:48:19] Speaker A: I mean, like, now could you I maybe another acquisition be announced on stage? But I could see that, like I could see lots of different things that could happen in the security space.
[00:48:27] Speaker D: I still think it's going to be something with security and AI, whether it's maybe more around like the actual communication of like and the guardrails of the AI and like giving people more guardrails and more guardrail options.
[00:48:47] Speaker A: Sounds great. Any honorable mentions you guys want to bring up? I will, I'll put the industry verticalization.
Industry verticalization for AI LM models either fine tuning, fine tuning marketplace or special models for specific industries, industry or use case.
Okay, any other honorable mentions?
[00:49:14] Speaker C: Yeah, I think that they're going to announce a.
A personal assistant for the Google Workspace productivity suite, but I wasn't sure if they actually covered that in this conference or not so.
[00:49:25] Speaker A: They do, they do cover workspace stuff. It's sort of a weird like oh, by the way we just goog Google mail.
[00:49:32] Speaker C: Yeah.
[00:49:33] Speaker D: Yeah. I was also wondering if like I don't think phones or anything but like some sort of like AI more like Gemini integration into the.
[00:49:41] Speaker A: They won't really cover Android stuff that's typically done at IO.
[00:49:45] Speaker C: Yeah, yeah.
[00:49:46] Speaker D: Okay. That's where I wasn't sure like where the lines were.
[00:49:48] Speaker A: They might, I mean they might mention something around Gemini and that'll get. You'll see it in Android as well, but it's typically not an Android specific announcement. Okay, that's like, that's. That's the whole build versus ignite conversation. Yeah, yeah, that's a Google I O which is their developer conference where they talk about Amazon or they talk about all of the phone stuff and the new version of Android, et cetera and whatever lovely delicious dessert it's named after. If they still do that. I don't even know because I don't have Android.
[00:50:15] Speaker C: Yeah, that's what I do.
[00:50:19] Speaker D: I don't think they formally do it, but I think it's still like one of those.
[00:50:22] Speaker A: I think internally they probably still do the code names but they just don't publicly announce it. All right.
[00:50:26] Speaker D: I mean everyone know they get leaked.
[00:50:29] Speaker A: Let's see. Did I have any other honorable mentions? I wanted to bring up AI suggested multi cloud tools which I didn't. I made that bet before, not won that one. So multi cloud tooling, it's possible I could see it.
[00:50:43] Speaker C: I was, I was stoked to find out it's a part of the security offerings.
[00:50:46] Speaker A: Yeah, yeah. So yeah, I think that as best as my crystal ball is for Google, I think that's a pretty good. Yeah, I think. All right, so then the question now, Matt you get a pick how many times I can say AI or ML on stage. You get first pick.
[00:51:03] Speaker D: 52.
[00:51:05] Speaker A: 52 times. This is both the keynote and the developer keynote.
Exclude the partner keynote.
I'm gonna go 97.
[00:51:16] Speaker C: Going big.
[00:51:16] Speaker D: I was actually.
[00:51:18] Speaker A: And Ryan, what do you think?
[00:51:21] Speaker C: It's closest without going over.
[00:51:23] Speaker D: Right?
[00:51:23] Speaker C: That's the rules.
[00:51:24] Speaker A: Yeah. So you go one if you wanted to because price is right rules.
[00:51:26] Speaker C: Yeah, I, you know, because my number was going to be lower than 52, so I am going to go one.
[00:51:31] Speaker A: All right, good call.
Might be the winning play right there.
All right, well, let's get into the things they announced that you guys did not predict, which is good.
Google is really nibbling on the edges of backup and disaster recovery this month with another feature this week, which I think is maybe a sign that ransomware is still a big problem for people or a concern.
Backup Vault was announced last year as a powerful storage feature available as part of Google Cloud Backup and doctor services. The point is to secure backups against tampering and authorized deletion and integrate with Security Command center for real time alerts on high risk actions. But to further support security needs, they're deepening the integration between Google Backup and Dr. And Security Command Center Enterprise. This includes new detections, including threats to the backup vault itself and end to end workflows to help customers protect their backup data more easily. Nice. That's good.
[00:52:19] Speaker C: Yeah, I mean I think not only is ransomware still a big issue, but also it's hit the compliance round. So there's like, it's a question that comes up all the time in any kind of security audit or attestation or even if like a customer walk through. So it's, it's definitely an issue that's in the front of people's minds and something that's really annoying to fix in reality. So this is great.
[00:52:45] Speaker D: Specifically like offline backups that are immutable because so many people have everything just in the cloud. Like yeah, we have Dr. In the other region and it's S3 applications, the other region or GZRS or whatever the extra letters are on Azure, you know, for you know, perpetual backup. But it's all online. So really the immutable piece is what I feel like is really the key here with a lot of this.
[00:53:11] Speaker C: Oh yeah, definitely. And it's, you know, it's just such a operational challenge. Right. With that, that immutable or you know, the write once read many sort of strategy. It's, it's not as easy. It's not the easy button that we've been, you know, privy to previously, then just dump it all on an S3 bucket.
But now you actually have to manage it and make sure the access is there and you can't fix it if you blow it up.
[00:53:37] Speaker A: Another thing that people think are dead but still kicking like a zombie is the mainframe. Yeah, and Google has a mainframe announcement this week, which is always weird when cloud provider has a mainframe announcement. Google and a company I've never heard of called Mlogica have announced an expanded partnership focused on accelerating and de risking mainframe application modernization, combining mLogical's Liber M automated code refactoring suite available at the Marketplace, Google Cloud Dual run for validations and de risking offering a validated modernization path to their joint customers.
Liber starm provides automated assessment, code analysis, dependency mapping and code transformation capabilities and supports multiple target languages and platforms, including a crucial foundation for refactoring your projects. Google Dual run, which I did not know existed, enables a simultaneous operation of mainframe and cloud applications in parallel, letting you compare and validate refactored applications before cutting over. This, along with powerful testing capabilities, enables a controlled phase transition, minimizing business disruptions and substantially reducing the risk inherent in large scale mainframe modernization projects.
[00:54:41] Speaker C: I mean, that's pretty smart because a lot of people refuse to get off of mainframes because of the perceived risk.
[00:54:46] Speaker A: Right?
[00:54:46] Speaker C: And they don't want to, even if it's just calculated differently or more correctly, you know, they still don't want that computational difference to be in there. So providing the sort of side by side comparison of the results is pretty smart.
[00:55:00] Speaker A: I mean, they could expand this beyond mainframe. There's a lot of business applications that, you know, people want to run parallel and. But building parallel pipelines is expensive and hard. Like there's lots of use cases for this.
[00:55:11] Speaker C: I wonder if we'll see a lot more of these announcements.
[00:55:14] Speaker D: I was just saying the dual run feature I think is really cool and because I've helped customers do these migrations before and it's like, yeah, it's just a pain to do. And at one point you're like, we're going to hit the big button and cut over and make sure and hopefully everything works, otherwise we're in deep trouble.
[00:55:30] Speaker C: Yeah, I wonder if we're going to see a lot more of these announcements. As you know, like the code modernization and AI sort of takes hold of that. Right? Because we know that some of the older languages there's. The brain drain has happened as people have retired and people don't know some of these things anymore. But AI can be trained on it pretty simply and you just, you know, generate new ones. Or maybe it'll even generate these modernization paths.
[00:55:58] Speaker A: Well, Google has a Geek out article for us guys. I always love these. They have another update on Colossus. Colossus is their expansion of their foundational distributed storage system, used to be called GFS or Google File System.
Basically this handles all of the object storage and file storage that Google offers to their cloud customers and has an easy to use programming model that's used by all Google teams. Moreover, does all this will serve the needs of products with incredible diverse requirements, be it scale, affordability, throughput or latency. And they gave some examples. They example application BigQuery scans hundreds of kilobytes tens of hundreds expected performance terabytes per second cloud storage standard kilobytes tens of megabytes hundreds of milliseconds expected performance Gmail messages tens of milliseconds of performance Hyper Disk reads less than a millisecond of performance and YouTube storage is in megabits and IO sizes but seconds in expected performance.
Basically the Colossus metadata Sorry, the Colossus GFS system is model is modeled to append only storage system that combines file systems familiar programming interface with scalability of object storage. Under the hood, the Colossus metadata service is made up of curators that deal with interactive control operations like file creation and deletion, and custodians which maintain the durability and availability of data as well as disk space balancing. Colossus clients interact with the curators for metadata and then directly store data on D servers which host either SSD or HDDs or hard disks.
[00:57:19] Speaker C: Basically.
[00:57:20] Speaker A: It's also good to understand that Colossus is a zonal product. They build a single Colossus file system per cluster, an internal building block of a Google Cloud zone, and most data centers have one cluster and thus one Colossus file system. Regardless of how many workloads run inside the cluster, many Colossus file systems have multiple exabytes of storage, including two different file systems that have an excess of 10 exabytes of storage each. And all I can say is someday there will be a massive Colossus related outage for a lot of customers potentially. I bet they have really good testing on this thing.
[00:57:46] Speaker D: Yeah, I'm just still remembering the S3 outage where the guy rebooted too many servers at once.
[00:57:52] Speaker A: Yep, demanding applications also need large amounts of iops and throughput. In fact, some of Google's largest file systems regularly exceed read throughputs of 50 terabytes a second and write throughputs of 25 terabytes per second. This is enough throughput to send more than 100 full length 8k movies every second.
[00:58:07] Speaker D: Wow. Wow.
[00:58:08] Speaker A: Their single busiest cluster does over 600 million IOPS combining between, combined between read and write operations. And previously when they talked about Colossus, they talked about how they place the hottest data on SSDs and balance the remaining data across all the devices in the cluster. This is more pertinent today as over the years the SSDs have gotten more affordable but still pose a substantial cost premium over blended fleets of SSD and HDD drives. To make it easier for developers, they have a L4 distributed SSD caching layer with dynamic pixel data that is most suitable for ssd. So the developers don't even have to think about the tiering. Take that Amazon. And overall this is a great deep dive into a bunch of other nuance and details of L4 caching, et cetera that we won't talk about here. But I love these articles because they just help you understand how your cloud is built.
[00:58:48] Speaker C: Yeah. And it, you know, like this is how you build systems at scale.
[00:58:52] Speaker A: Right?
[00:58:52] Speaker C: Because we don't use Colossus directly, we use services that are run on Colossus. And so like the teams, all those individual teams that are writing those services now have just a common framework, a common system. The enablements of like you said, not having to, to really think about the tiering, it's just all built into the service and allows you to focus on, you know, your area of the service that matters the most, which is the customer experience. And I, you know, like I love to see companies sort of develop and then also tout their back end systems like this because it's super awesome. Especially with you know, metrics at this scale.
[00:59:26] Speaker A: Yeah, I mean I love when you shared with me secret NDA data from Yahoo days scaling Yahoo Mail. Yeah, like that's just fun. I mean those are all old, old technologies now.
[00:59:36] Speaker C: Yeah, I mean it's, it's not really NDA because there was, you know, a whole bunch of tech posts about it. But yeah, no, I mean this is near and dear to my heart just because of that, you know, when we're trying to adopt a new Ceph service and trying to build basically an internal service like this. So this is fantastic.
[00:59:52] Speaker A: Yeah, it is really cool the scale they're dealing with here too. I mean we talk about S3 during PI Day, you know, and how much it's doing. Like this is there is Google's version of what Amazon's built. Now the thing that Amazon and Google can differ on is that this is a big massive service that lots of things use. And the fact that it's one cluster per zonal region is kind of shocking to me. But I mean it makes sense when you think about the metadata and the direct write to the D server. So it makes some sense that it's not required to be, be, you know, more than one cluster because again you're, you're separating the loadout in a fully distributed model. So it makes sense that that's not a big issue. It just, you know, when you say it that way, I don't know if I would have worded it that way. Like, you know, I think that's where Amazon kind of did a little better talking about Micro was it talked about S3, talked about like, you know, all these little teeny tiny pieces of your files floating around and.
[01:00:43] Speaker C: Yeah, yeah, oh yeah, yeah. You know, the sharding of data around multiple places.
[01:00:47] Speaker A: And so like, you know, yeah, we have these like tons of clusters and your data, you got all this durability because of this. And so there's some things in how you market this is important. But basically Amazon has something similar, Google has something similar. I'm sure Azure pretends they have something with their Azure San that's similar to.
[01:01:03] Speaker D: This, but just netapps under the hood, it's fine.
[01:01:05] Speaker A: Yeah, it doesn't add up all the way down at Azure, but yeah, this is cool. So this is definitely geek out if you're into these kinds of things. Read the more details. I just try to gloss over a bunch of the high level points, but definitely interesting.
[01:01:19] Speaker D: It really just helps also build confidence. I feel like in the clouds, you know, hyperscalers. While you don't think about it, you know, they are thinking about all these ridiculously complex problems so you know, you can go focus on what your business actually needs, which is nice. So when somebody goes and says hey, are you sure we really should move here? Can Google handle our, you know, one terabyte of data? And you're like, like here's an article where they handle billions of times more than that. Your 1 TB is nothing to them.
[01:01:47] Speaker C: I just want dev teams to enable snapshots in our resources. I mean I love hearing services like this because this is what, you know, I'd rather be working on sort of chasing Teams down for compliance reasons, you.
[01:02:00] Speaker A: Chose to move to security.
[01:02:01] Speaker C: I know.
[01:02:03] Speaker A: I had just got my risk register jury tickets today. I was like, now it's gonna be Ryan complaining to me about these now.
[01:02:10] Speaker C: Great.
[01:02:12] Speaker D: Maybe they'll be more real now because Ryan's doing.
[01:02:15] Speaker C: Oh, no, no, no, no.
[01:02:16] Speaker A: I mean, hey, Ryan has already gone and fought the good fight with our compliance team on many of them for me, which I really appreciate that he's done that work to show him like here's how it's mitigated and not the risk that you think it is. And he's done a really good job, like de risking a bunch of them. But they still exist. I still have to solve them. I just don't solve them tomorrow.
[01:02:32] Speaker D: Yeah, right. They are now no longer critical as you have seven days, you have 30 days or whatever.
[01:02:38] Speaker A: Yeah.
All right. In our Last Google Story, AI assisted BigQuery data preparation now generally available. Surprised they didn't save this one for Google Next. Actually.
BigQuery data preparation is now generally available and also it now integrates with BigQuery pipelines, letting you connect data ingestion and transformative tasks. So you can create end to end data pipelines with incremental processing, all in a U5 environment. Features include comprehensive transformation capabilities, data standardization, automated schema mapping, AI suggested join keys for data enrichment, visual data pipelines, data quality enforcement with error tables, and streamline deployment with GitHub integrations we talked about last week.
[01:03:11] Speaker C: Automated schema mapping man, it's probably my biggest life work improvement.
[01:03:17] Speaker A: Could Athena get that please?
[01:03:19] Speaker C: Yeah, right.
[01:03:19] Speaker D: Yes.
[01:03:21] Speaker A: It's the bane of my existence.
[01:03:22] Speaker D: Even just for the Azure, you just for the AWS tables, the ones that AWS can you just automatically know those?
[01:03:30] Speaker A: Yeah, if AWS you're generating this data that I'm now indexing. Athena, please just provide me the schema.
[01:03:36] Speaker C: Out of a library in an accurate one, please.
[01:03:39] Speaker A: Yeah, yeah, yeah.
[01:03:40] Speaker D: Not oh, we've added these rows to ALB flow log or ALB access logs like no, no, just know what it is and move on with my life.
[01:03:48] Speaker A: If they could just use their Amazon Cube or this Nova thing to go update all the documentations that has the same examples because the schemas examples they have are wrong all over the place. You have to figure out which one's the most current to get the right version again, a library would be great. Amazon, if you're listening to us.
All right, let's move on to Azure.
If you're using Microsoft fabric surprise, you get a new bill.
They're going to be charging you for those SQL backups. Previously, your fabric capacity based billing model included compute and data storage. And by default the system provided a full weekly backup differential backup every 12 hours and a transaction log backup every 10 minutes. But after April 1, 2025, backup storage will also be billed and that exceeds the allocated DB size. So if your database is 100 gigabytes and your backups add up to more than that, you will now be paying for that. Which is nice, I guess, but I'm a little perplexed because. Can I turn this off? Can I change the retention?
[01:04:42] Speaker D: That's what I was wondering.
[01:04:43] Speaker A: Yeah, because I mean like, because you know, a full backup alone, probably compressed, is probably going to be at least half the size of your database. So then you're doing differentials every 12 hours and you're doing a 10 minute transactional backup. If it's a busy database, you're gonna be using a lot more storage than what the original database is on a weekly basis.
[01:05:00] Speaker D: Probably what happened was they realized how much more storage this is actually using and they were like, guys, this is.
[01:05:07] Speaker A: A revenue leak problem, we need to solve it.
[01:05:09] Speaker D: Right?
[01:05:09] Speaker C: Yeah.
[01:05:09] Speaker D: Yeah.
[01:05:10] Speaker C: I'm kind of hoping that was an April 1st announcements that this is gonna be a joke tomorrow, but I don't think so.
[01:05:18] Speaker D: No, I think this was a few days ago.
[01:05:19] Speaker A: This is posted on March 27th. Yeah.
[01:05:21] Speaker C: Oh, but it's in the announcements after April 21st.
[01:05:24] Speaker A: Oh yeah, no, the billing starts after April 1st.
[01:05:26] Speaker C: Yes, that's true and I realize that it's not going to be that, but it's just like this is terror. I don't like these kinds of announcements because there's no value add to the customer. Right.
[01:05:35] Speaker A: You know, well what I, what I find annoying. So this, this blog post was posted on March 27, 2025. You had four days and the timeline they said was there was no change until March 21. And then from March 21 to April 1 they were enabling the viewing so you could see what it was going to cost you. But non billable. But then they didn't release this till the 27th, but it's effective in four days. Like what? Yeah, like that's ridiculous. Like at least you know, post this blog post and give yourself two weeks to figure it out. Like, or a month would be more ideal, but you know, it's barely enough time.
[01:06:04] Speaker C: Just turn off all your backups.
[01:06:06] Speaker A: Right. Which maybe you shouldn't turn off.
[01:06:10] Speaker D: I wonder if they went to all the customers that were any real amount behind the scenes.
[01:06:14] Speaker A: I mean, I hope so. I hope that's the case.
[01:06:15] Speaker D: And like, we're like, hey, guys, just so you know, your bill's about to double. Good luck.
[01:06:19] Speaker A: This is how we learned Matt doesn't use fabric.
I'm sure. I'm surprised because to do any reporting on Azure, you have to have fabric.
[01:06:28] Speaker D: It sounds like I will say fabric is my weak point that I need to sit down and like, really understand. And I just haven't spent the time.
[01:06:37] Speaker A: On it because we've had a crash course in it. This is an area that Ryan and I can actually be educational to you because we had to enable fabric and to use Power BI for a product. And so we connected Power BI to gcp and I can tell you all about it.
[01:06:50] Speaker C: To be fair, I can mostly just tell you what not to do.
[01:06:54] Speaker D: Use fabric.
Just use fabric. Start at that level.
It just feels like it's their catch all service for everything.
And I know some stuff they're building in in the future, which are pretty cool.
[01:07:09] Speaker A: But that's why I call Fabric their version of Snowflake, because that's exactly what it feels like. It's a. It's a catch. All of a ton of analytical AI stuff that you may or may not need, but you'll stitch it together into something like Snowflake uses.
[01:07:23] Speaker D: That's better. I understand about it.
[01:07:25] Speaker A: Yeah.
So this is where I wonder about security, Ryan.
So they're announcing alert Triage agents and Microsoft Purview powered by Security Popilot, which I'm like, cool. Microsoft says per their research that organizations are facing up to 66 alerts per day when it comes to purview DLP alerts alerts up from 52 in 2023, with teams only really able to review about 63% of the alerts. So when I math that that means that the SOC team is only handling half of those 52 alerts or 33 alerts this year. And given the sheer volume of data security alerts, it's no surprise that Microsoft says it's hard to keep up with these. Do you know how many alerts I get in OPS per day? Like 66. I dream of that.
[01:08:10] Speaker C: Yeah, I mean, I. I get you.
[01:08:13] Speaker A: Well, apparently Microsoft's trying to help you, but not me. Yeah, because they're giving you the help to help customers increase the efficiency of their data security programs. They're addressing key alerts and focus on the most critical data risks. Microsoft is thrilled to announce Alert Triage agents and Microsoft Purview data loss prevention and insider risk management or erm. These autonomous security copilot capabilities integrate directly into Microsoft Purview. Offer an agent managed alert queue that identifies DLP and IRM issues that pose the greatest risk to your business. Business.
[01:08:38] Speaker C: So I'd argue that the entirety of the cloud is to help you with your obs alerts.
[01:08:46] Speaker A: I don't know, I don't know by that, but okay, sure.
[01:08:49] Speaker C: Auto scaling instead of the disk filling up full of logs. Sure, sure.
[01:08:53] Speaker A: Okay.
[01:08:56] Speaker D: You clearly don't use Azure where things just hit alerts and it doesn't scale or it can't scale or it can never scale down if it scales up.
[01:09:06] Speaker C: I mean there is a harsh reality in these numbers. Right. Because it's these. I, I think they're right or optimistic.
[01:09:14] Speaker A: Right.
[01:09:15] Speaker C: In the number of alerts that I actually looked at.
And it's, you know doing something with DLP is really tricky because you don't want to get all up in users data but you have to sort of figure out how to make sure that you are protected from data loss. And so each one of these investigations for these alerts is a little, you know, it's, it's very time consuming. Time consuming and very manually done, you know with talking to people and making sure and.
Yeah.
[01:09:45] Speaker D: So you looked at Security Copilot.
[01:09:48] Speaker A: I mean they just announced it a couple weeks ago.
[01:09:49] Speaker D: No, Security Copilot's been out for a long.
[01:09:51] Speaker A: Oh yeah, sorry. They just announced some enhancements to it.
[01:09:53] Speaker D: So I'm just sitting here reading this going. Purview has multiple levels. Security Copilot when I looked at by itself was base. I think it was several grand a month at best. On the cheap end.
[01:10:07] Speaker C: Yeah.
[01:10:08] Speaker D: Security tools aren't cheap, just felt very, very expensive.
[01:10:12] Speaker C: And there's a billion of them and you need all of them to make up a narrative because security narratives aren't are, you know, are less about the secure environment that you're creating and more about the narrative of the secure environment that you're creating. Like you have to do a sales job every single time. So it's, it's tricky and you know, one of the options that people use is to use tools that everyone knows.
[01:10:36] Speaker D: Yeah.
Oh, I'm very familiar with it.
[01:10:43] Speaker A: All right. I have Oracle announcements for you this week.
[01:10:45] Speaker C: A bunch of them.
[01:10:46] Speaker A: A bunch. I know. Three, three lovely announcements beat Azure. Yeah, I know, it's a weird, weird week. OCI or Oracle Cloud is making available the Latest and greatest Nvidia GB300, NVL72 and Nvidia HGX B300 NVL16 with Blackwell Ultra GPUs providing early access to the AI acceleration capabilities of these new models. You get the GB300, the B300 in bare metal or you can use superclusters up to 131,000 Nvidia GB300 Grace Blackwell Ultra superchips as part of a rack scale Nvidia GB300 NVL72 solution. Which made me curious, how much does 131,000 Nvidia GV300 Grace Blackwell Ultras cost? And I can't answer that question for you as it was not an option in the pricing calculator. However, I was able to choose a GPU optimized instance which I picked the one option which was the BM GPU GB204 with 4 GPUs and 700 gigs of memory running autonomous Linux for the low low price of $857,088 in monthly on demand costs.
I have no idea. I think it's so four GPUs so I mean if I were to do the math of four GPUs into the GV300, that must be very, very expensive to get a super, super cluster of that scale and size. But yeah, that one I was like, wow, that is more than I owe on my house, right? Yeah, by a lot.
[01:12:09] Speaker C: Never mind like making services was too two pizza teams. Like they're just burning money for two pizza teams, right?
[01:12:16] Speaker A: Like I'm talking about CFO is having a bad day. Get that bill right.
I, I also did not. I could not figure out how to toggle. The Oracle calculator is terrible. I don't know if you're a lieutenant ever, but there's no easy way to choose between like on demand versus committed, you know, resources easily that I could see in this particular instance type. So I only have the on demand cost, but I assume you get some pretty hefty discounts. I hope at least half of that for 400,000 a year.
But anyways, I hope you get, you know, multiple engineering resources out of those GPUs to justify the cost.
[01:12:49] Speaker D: I want to be somewhere where I can turn to my CFO and say, do you for this project, I want to, I just want to launch it once.
[01:12:58] Speaker A: I just think it'll be fine once. I want to, I want to pay for one second of it, please.
[01:13:01] Speaker D: Yeah, like that's it. I just want to do the terraform apply, watch it launch, look at the console and go, huh. There goes all that money.
[01:13:08] Speaker A: I want to run Windows on it so I can open it up and see Task Manager and see all the CPUs just scaling off the screen.
[01:13:14] Speaker C: I just want to make it the default option on the thing that I deploy. So if you don't actually read the documentation, it spits up one of these guys.
[01:13:22] Speaker A: Yeah, I was, I was pleased to see that if I chose something other than autonomous Linux, the price didn't change. So if I chose Ubuntu or OpenSuse the same price, the licensing of the Linux is included in this.
[01:13:33] Speaker D: I wonder what the licensing for Windows of that would be.
[01:13:36] Speaker A: Oh, I don't know. I mean it doesn't have that many actual CPUs. That's the saddest part about it. So you're paying for the GPUs.
[01:13:44] Speaker D: That's true.
[01:13:45] Speaker A: So for in instances that are more R cost model, the new More reasonably priced E6 standard bare metal and Flex virtual machine instances are now available. Powered by the 5th gen AMD Epyc processors. OCI says they're among the first cloud providers to offer them. I checked this up. Google was the first. Oracle is the second.
[01:14:01] Speaker D: So so AWS announcement next week.
[01:14:05] Speaker A: No AWS or Azure yet on this one I had to check. These are the turn based AMD chips. They're not yet for either of those two cloud providers. Oracle is promising a performance of two times that of the E5 at the same price, which is not a bad deal. This has a 2.7 GHz base frequency with a max boost up to 4.1 GHz based on the Zen 5 Nvidia or sorry AMD architecture. You can configure the virtual ones for 1 to 126o CPUs and up to 3 gig or 3 terabytes of bare metal memory or 1454 gigabytes of virtual machine memory. There are sponsor limitations in disk throughput and network bandwidth based on if it's bare metal or not.
I didn't learn the pricing on this one by the way, but it's much cheaper than the one above.
[01:14:48] Speaker C: It's probably still very expensive.
[01:14:50] Speaker D: I'd hope so.
[01:14:51] Speaker A: I mean, I doubt it. It's AMD Epyc. It's got to be at least 10% cheaper than an Intel.
[01:14:56] Speaker C: That's probably. That is true. It's just a lot of memory but.
[01:14:59] Speaker A: Yeah, I mean the memory will cost you. I'm sure if you do bare metal like that'll cost you a bit of money. But again, you're probably not talking six digits. Maybe five digits.
[01:15:08] Speaker C: Yeah, I would agree. Yeah, I'd still get in trouble for running anything on this.
[01:15:15] Speaker A: Well, I mean, you'd be fine if you did the virtuals and did small, but yeah, if you're trying to run the full box, you'd probably get in some trouble.
[01:15:21] Speaker D: It's interesting that Oracle is the second provider that in it and my next question is how many regions is it actually launched in?
[01:15:29] Speaker A: It's in a surprising number of regions for day one. Funny you should ask that. It's in the article.
It is available in U.S. east Ashburn, U.S. west, Phoenix, U.S. midwest, Chicago, Germany, Central Frankfurt and UK South London.
[01:15:42] Speaker D: Okay, so five.
[01:15:43] Speaker A: Hey, it's better than one.
[01:15:45] Speaker D: Yeah, but like how many regions does Oracle have? I feel like they must be up to like 90 trucks around the world.
[01:15:52] Speaker C: Yeah, but they're small trucks so they can't feed everyone.
[01:15:56] Speaker A: Yeah, you can't get a lot of them. Let's see bare metal servers. See if I can find the cost of this for you guys. Real time, best podcast content ever. When you do these real.
[01:16:07] Speaker D: There are a lot of dots in the Oracle public region map.
[01:16:12] Speaker A: There are a lot of dots. That is correct.
How are you in the right box? Standard. This is an E5. Right. Epyc9J E5 based standard compute processor. Andy. Epic9J14 base 2.4. Nope, this is the right box.
[01:16:25] Speaker C: E6 is the. This is compared to the E5.
[01:16:28] Speaker A: Oh, that's right. E6 is the new one. Okay. Yeah, here it is. $10,285 for a bare metal running autonomous Linux. So that's actually not that bad.
It does jump up to $27,000 if you go for Windows.
[01:16:43] Speaker D: Wow, thank you, Windows.
[01:16:46] Speaker A: Yeah. So not bad. I only added 100 gigs of disk space because who needs more than that? But you're using Windows 25 cents of data in there. Windows will cost you a bunch of money.
[01:16:56] Speaker D: Yeah, Windows by default you have to have more than 100 gigs. The OS is that much.
[01:17:04] Speaker A: I don't. I guess I don't. So a capacity reservation didn't change the price, which is sort of weird.
Oh, I see. See? Do you tell it how many? Okay, I don't. Oracle pricing.
[01:17:14] Speaker C: Weird.
[01:17:15] Speaker A: Real time podcast follow up over.
[01:17:17] Speaker D: Yeah.
[01:17:17] Speaker A: All right, move on to the last Oracle story. Oracle is apparently under fire for potential security breaches. The first one is related to Oracle Health, which is their acquisition of Cerner. The breach impacts patient data and apparently they're blaming on an old legacy server not yet migrated to Oracle Cloud. Way to pass the buck. And then the Other breach may be on Oracle Cloud and the Oracle is being a little cagey about this one. A hacker going by rose87168 posted on cybercrime forum offering the data of 6 million million oracle Cloud customers, including authenticated data and encrypted passwords. Several Oracle customers have confirmed that the data appears genuine, but Oracle has stated that there has been no breach and published credentials are not from the Oracle cloud. So where are they from? Oracle? There's a cyber security expert, Kevin Beaumont writes, this is a serious cybersecurity incident which impacts customers and a platform managed by Oracle. Oracle are attempting to wordsmith statements around Oracle Cloud and use very specific words to avoid responsibility. And this is not okay. And my point on that is Oracle cloud can't be unbreakable if it was hacked.
[01:18:14] Speaker D: Yeah, they didn't break it. It's still running.
[01:18:17] Speaker A: True, true.
[01:18:18] Speaker C: Yeah, it's just hemorrhaging your user data. No problem.
[01:18:21] Speaker D: Yeah, no problem.
[01:18:22] Speaker A: Yeah, no issue. All right, gentlemen, we have made it through another fantastic week here at the Cloud Pod. We are ready for Google Next. I'll see you next week, Ryan, in Vegas. I will not see you, Matt, but we'll think about you remotely.
Not really, but. Well, you know, maybe we'll think about you.
[01:18:38] Speaker D: We might just send me pictures.
[01:18:39] Speaker C: We might drunk dial you.
[01:18:40] Speaker A: Yeah, we might. That might happen because we have, we have some of your other friends that you know that will be there with us. So we can definitely, you know, include you in the trunk dialing.
[01:18:47] Speaker D: You should. I'm looking forward to it. If not, I'll just randomly call you guys.
[01:18:51] Speaker A: Yeah, so you get a week off. Ryan and I are busy handing out stickers, hopefully. But you have a week off, so we'll see. We will record the next episode. Not during Google Next week because we'll be busy, but the following week will be our recap show. It'll be a Google wall to wall show or we assume, unless they announce nothing. And then we'll replan that show. But I will have one and Ryan will have one.
But we will definitely talk to you guys here in two weeks when we're back from our little break as we go to Google Next. So we'll see you on the next show. Have a good time in the Cloud. Anytime.
[01:19:20] Speaker C: Bye, everybody.
[01:19:24] Speaker B: And that's all for this week in Cloud. We'd like to thank our sponsor, Archera. Be sure to click the link in our show notes to learn more about their services.
While you're at it, head over to our
[email protected] where you can subscribe to our newsletter, join our Slack community, send us your feedback and ask any questions you might have. Thanks for listening and we'll catch you on the next episode.