 Live from the Moscone Center, it's theCUBE. Covering AWS Summit San Francisco, 2018. Brought to you by Amazon Web Services. Well, we're welcome back to our exclusive CUBE coverage here at AWS Amazon Web Services Summit 2018 in San Francisco. I'm John Furrier with my co-host, Stu Minim. We have a special guest, we have an influencer, authority figure on AWS, Corey Quinn, editor of Last Week in AWS. Also has got a podcast called Screaming in the Cloud. Screaminginthecloud.com just launched. Corey, great to have you on. Thanks for joining us. No, thank you for letting me indulge my ongoing love affair with the sound of my own voice. I appreciate it. Well, we'd love to have you on. And again, love the commentary on the keynote on Twitter. A lot of action, we were in the front row, kind of getting all the scene. Look it, if you're going to write the newsletter next week for what happened this week, if this week was last week, tomorrow, next week, what's your take on this? Because again, Amazon keeps pounding the freight train that's just the cadence of AWS denouncements. But they're laying it out clear. They're putting up the numbers, they're putting out the architecture, they're putting out machine learning. It's more than developers right now. What's your analysis, what's your take of what's happening this week? I think that certain trends are continuing to evolve that we've seen before. Where it used to be that if you're picking an entire technology that you're going to bet your business on and what you're going to build on next. It used to be, which vendor do I pick? Which software do I pick? Now, even staying purely within the AWS ecosystem, that question still continues to grow. Oh, so I want to use a database. Great, I have 12 of them that I can choose between and whatever I pick, the consensus is unanimous, I'm wrong. So there needs to be, I still think there needs to be some thoughtful analysis done as far as are these services solving different problems? If so, what are the differentiating points? Right now, I think the consensus emerges that when you look at a product or service offering from AWS, the first reaction all of us feel is to some extent confusion. I'm lost, I'm scared, I don't really know what's going on. And whatever I'm about to do, I feel like I'm about to do it badly. Yeah, scale's the big point. I want to get your reaction. Matt Wood, Dr. Matt Wood, Cube alum, many times, he nailed it. I thought when he said, look at machine learning and data analysis was on megabytes and gigabytes. They're offering pedoflop level compute, high performance, and then Werner Vogels also said something around the services where you can open things up in parallel at scale. So what's your reaction to that? As you look at this, whoa, I got a set of services that can launch in parallel in the scale of leveraging that pedoflops. I mean, this is kind of like the new compute model. Your reaction, is it real? Are customers ready for it? I mean, where are we in that evolutionary customer journey? Are they still cavemen trying to figure out how to make fire and make a wheel? I mean, where are we with this? I think that we see the same thing continuing to emerge as far as patterns go where they talk about, yes, there's this service, just start using it and it scales forever. And that's great in theory. But in practice, all of the demos, all of the quick starts, all of the examples, paint by numbers, examples that they'll give you tend to be at very small scale. And yes, it works really well when you have effectively five instances all playing together. But when you have 5,000 of those instances, a lot of sharp edges start to emerge. Scale becomes a problem. Fail overs take far longer and let's not even talk about what the bill does at that point. Additionally, once you're at that point, it's very difficult to change course. If I write a silly blog and effectively baby seals get more hits than this thing does, it's not that difficult for me to migrate that. Whereas if I'm dealing with large scale production traffic that's earning me money on a permanent basis, moving that is no longer trivial or in some cases feasible at all. Yeah, Corey, how does anybody reasonably make a decision as to how they're going to build something because tomorrow, everything might change and said, oh, okay, great. I had my environment. I kind of built my architecture a certain way. Oh wait, there's a new container service. Oh, and start building up. Oh wait, now there's the orchestrated version of that that I need to change. Oh wait, now there's a serverless built way that kind of does it a similar way. So it seems like it used to be the best time to do thing would have been two months ago, but now I should do it now. Now the answer is the best time for me to do things would be if I could wait another quarter, but really I have to get started now. I tend to put as much on future Corey as I possibly can. The problem is, is that at one time I could have sat here and said the same thing to you about, oh, virtualization is the way to go. You should migrate your existing bare metal servers there, and then from virtualization to cloud and cloud to containers, then containers to serverless. And this narrative doesn't ever change. It's oh, what you're doing is terrible and broken. The lords of thought have decreed that now it's time to do this differently. And that's great, but what's the business use case for doing it? Well, we did this thing that effectively people get on stage at keynotes and make fun of us for now, so we should really change it. Okay, maybe, but why? Is there a business value driving that decision? And I think that gets lost in the weeds of the new shiny conference ware that gets trotted out. Well, I mean, Amazon's not, I mean, they're pretty being forthright. I mean, you can't deny what Intuit put out there today. The Intuit head of machine learning and data science laid out old way, new way. Classic case of old way, new way. Eight months, six to eight months, ton of cluster, you know what going on as things change. They're just data scientists and not backend developers. They went to one week, nine months to one week. That's undeniable, right? I mean, how do you, I mean, that's a big company, but that seems to be the big enchilada that Amazon's going for, not the pockets of digital disruption. You know what I'm saying? So it's like, how do you square that out? I mean, how do you think about that? Cloudability had a great survey that they released the results of somewhat recently where they were discussing that something like four or five of the, I'm sorry, 85% of the global spend on AWS went to four or five services that all have been around for a long time. RDS, EC2, S3, EBS, data transfer. And so as much as people talk about this and you're seeing pockets of this, it's not the common case by a wide margin. People don't get up on stage and talk about, well, I have these bunch of EC2 instances behind a load balancer storing data on S3 and that's good enough for me because that's not interesting anymore. People know how to do that. Instead, they're talking about these far future things that definitely add capability but do come at a cost today. It's a classic headroom. It's like give us some headroom but at the end of the day, it's EC2, S3, Kinesis, Redshift, bunch of services at SQS that seem to dominate. Question I want to ask you is that they always flaunt out the, every year it changes. Kinesis was at one point the fastest growing service in the history of AWS. Now it's Aurora. Do you, I made a prediction on the opening that a SageMaker will be the fastest growing service because it just seems to be so much interest in turnkey machine learning. It's hard as you know what to do it. Your thoughts on SageMaker. In one of my issues a few weeks back, I wound up asking, so who's using SageMaker and for what? And the response was ridiculous. What astounded me was that no two answers were alike as far as what the use case was but they all started the same way. I'm not a data scientist but. So this is something that's becoming bad. What does that mean to you? What does it tell you? It tells me that everyone thinks they're unqualified to be playing around in the data science world but they're still seeing results. Look, Corey, I wonder because think back a few years ago that was part of the promise of big data is we have all this data and we're going to be able to have the business analyst rather than some PhD sort this out and machine learning is more right. We want to have these tools and we want to democratize data. Data's the new bacon, data's the new oil, data's the new everything. So machine learning, do you think this is all vapor and promise or do you think it's real? I think big data is very real and very important. Ask anyone who sells storage by the gigabyte and they will agree with me. In practice, I think it's one of those areas where the allure is fascinating but the implementation is challenging. Okay, we have history going back 20 years of every purchase someone has ever made in our bookstore. That's great. Why do I still wind up getting recommendations? Well, yeah, and I guess I want to talk about it. I see it more as everything that was big data is now kind of moving to the ML and AI statements because big data didn't deliver on it. Will this new wave deliver on the promise of really extracting value from my data? And it's things like it's live data. It's doing things now with my data, not the historical, lots of different types of data that we were trying to do with the Hadoops of the world. Gotcha. I think that it's a great move because either yes it will or no it won't but if it doesn't, you're going to see emergent behaviors of so why didn't it work? Well, we don't understand the model that this system has constructed. So we can't even tell you why it's replacing the character I with some weird character that's unprincipled. So let alone why we decided to target a segment of customers who never buys anything. So it does become defensible from that perspective. Whether there's something serious there that's going to wind up driving a revolution in the world of technology, I think it's too soon to say and I wouldn't dare to predict but I will be sarcastic about it either way. Okay, well, let's get sarcastic for a second. I want to talk to you about some moves other people are making. We'll get to the competition in a minute but Salesforce acquired MuleSoft. That got a lot of news. We were speculating on our studio session this week or last week with the CEO of Rubrik that it's great for Salesforce that can bring yes, you know, structured data in on-prem and in cloud because Salesforce is one big SaaS platform. Amazon is trying to satisfy business through the cloud. So, but one of the things that's missing from MuleSoft is the unstructured data. So the question for you is, how are you seeing and how is your community looking at the role of the data as a strategic asset in a modern stack one, both structured and unstructured data? Is that even happening or is it more like, well, we don't even know what it means? Your thoughts. I think that there's been a long history of people having data in a variety of formats and being able to work with that does require some structure. That's why we're seeing things emerging around S3's increasing capabilities of being able to manipulate data at rest. We're seeing that with S3 and Glacier Select. We're seeing it with Athena, which is named after the goddess of spending money on cloud services. And there's a number of different tooling options that are, okay, we're not going to move three exabytes of data in so we have to do something with where it is. As far as doing any form analysis on it, there needs to be some structure to it in order for that to make sense. From that perspective, MuleSoft was a brilliant acquisition. But the question is, is what is Salesforce going to do with that? They have a history of acquisition, some of which have gone extremely well. Others of which we prefer not to talk about until like company. I mean, because back down the IDE thing too, how many IDEs does that Salesforce have now? I mean, it's a huge number. I'm sure at least three more since we started talking. Yeah, so Corey, you brought up money. So the trillion dollar price, what feedback are you getting from the community? There's always, well, I get an Amazon and then my bills continue to go, continue to grow. Same thing at Salesforce, by the way, if you use them. So, you know, there's always as you gain power, people will push back against it. We saw it with Microsoft, we saw it with Oracle. I hear it some, but it's not an overriding thing from when I talked to customers about Amazon. But I'm curious what you're hearing. Where are the customers feel they're getting squeezed? You know, where is it, you know, phenomenal? And, you know, what are you seeing kind of on the monetary side of cloud? In my day job, I solved one problem. I fixed the horrifying AWS bill, both in terms of dollars and cents, as well as analysis and allocation. And what astonishes me, and I'm still not sure how they did it, is that AWS has somehow put the onus onto the customer. If you or I go out and we buy $150,000 Ferrari, we wake up with a little bit of buyer's remorse of dear Lord, that was an awful lot of money. When you do the equivalent in AWS, you look at that and instead of blaming the vendor for overcharging, instead we feel, wow, I'm not smart enough. I haven't managed that appropriately. Somehow it's my fault that I'm writing that looks like a phone number of a check every month over to AWS. It creeps up on you. It does. It's the boiling of frog problem and by the time people start to take it seriously, there's a lot of ill will. There's a sense of our team is terrible and wasn't caring about this, but you don't ever cost optimize your way to success. That's something you do once you have something that's up and working and viable. It's not, you don't start to build a product day one for the least possible amount of money and expect to attain any success. Well, let's talk about that real quick in the segment because I think that's a really important thing is success is a double-edged sword. The benefit of the cloud is to buy what you need, get proof of concept going, get some flywheel going or whatever, virtuous circle of the application. But at some point you hit a tipping point, well, oh shit, this is working. And then the bill's huge. Better than over provisioning having a failed product. So where's that point with you guys of what we're customers? Is there like analytics you do? Do you, is that more of a subjective qualitative thing? You say, okay, are you successful? Now let's look at it. So how do you deal with customers? Because I can imagine that success becomes the opportunity, but also the problem. I think it's one of those you know it when you see it type of moments where if a company is spending $80,000 a month on their cloud environment and could be spending 40, that's more interesting to a company that is three people than it is an engineering team of 50. At that point, sorry, they're impressing more than office supplies every month. So that's not the best opportunity to start doing an optimization pass. More important that both of those scales to me has always been about understanding the drivers of it. So what is it that's costing that? Is it a bunch of steady state things that aren't doing work most of the time? Well, maybe there's an auto-scaling story in there. Maybe there's a serverless opportunity. Maybe no one's using that product and it's time to start looking at rolling it into something else. They left the lights on, right, so to speak. The servers are still up. Wait a minute, take them down. So writing code, analytics, is that the answer? All of the above. In a vacuum, if you spin up an instance today and Leah don't touch it again, you will retire before that instance does and it will continue to charge you every hour of every day. Understanding and being able to attribute, who spun that up, when was it done, why was it done, and what project is it tied to? Is it some failed experiment someone did who hasn't worked here in six months? Or is that now our master database? We kind of need to know in either direction what that looks like. All right, before we wrap, got to tell us what should we expect to hear from your podcast? Good question. My podcast generally focuses on one-on-one conversations with people doing interesting things in the world of cloud, which is big enough for me to get away with almost anything as far as it goes. It's less sarcastic and snarky than some of my other work, and more at the why instead of the how. I'm not going to sit here and explain how to use an API. There are people far better at that than I am, but I will talk about why you might use a service and what problem it purports to solve. All right, Corey, great to have you on. Screamingcloud.com. Screaminginthecloud.com. It's a podcast. Corey, thanks for coming on. Share in the commentary and insight on AWS, the how and the why, the cube breaking down, all the action here in Moscone, Western San Francisco, AWS 2018 Summit. Back after more with this short break.