AWS Made Easy
Search
Close this search box.

Ask Us Anything: Episode 6

With Steve Brain, Head of Technical Product & Portfolio CTO/EVP, Trilogy

In this episode, Rahul and Stephen interview Steve Brain, Head of Technical Product & Portfolio CTO/EVP at Trilogy. Steve has a wealth of experience in all things serverless. Rahul, Stephen, and Steve discuss a few announcements from AWS, and then have a discussion about serverless with #aws in 2022.

Latest podcast & videos

Summary

We begin the show by reviewing 3 AWS announcements, and then we have our discussion on serverless.

Announcement #1 – Amazon AppStream 2.0 adds larger instance sizes to the General Purpose instance family

The first announcement we discussed was about AWS AppStream2.0. AWS has added larger instance sizes to the General Purpose instance family. This means that even more tasks can be accomplished using AppStream, compared to needing on-prem hardware. Rahul mentioned that even the SolidWorks demo uses AppStream under the hood to quickly provision an instance. With AppStream, it makes it possible for IT to manage cheap thin clients with no proprietary data on board, and do the heavy computing in AppStream.

[1] : https://www.solidworks.com/

Announcement #2 – Amazon Personalize adds support for unstructured text in six new languages

The panel discusses AWS Personalize, and how text-based recommendation systems have empowered businesses to make better product suggestions to their customers, without needing to train and manage complex ML models.

Announcement #3 – AWS Cost Allocation Tag API

Rahul, Stephen, and Steve discuss the Cost Allocation Tag API, and the usage of tags for understanding distribution of costs in AWS accounts.

Discussing Serverless Architecture

We spent the remainder of the show discussing serverless architecture. Questions included:

  • What is “serverless”?
  • What are your thoughts on Aurora Serverless v1 vs v2?
  • How do you incorporate serverless into an existing application or product?

Kevin Duconge

Transcript

  1. [music]

  2. [00:10:18]

  3. Stephen

    All right. Hello, everyone and welcome to “AWS Made Easy”, episode…ask us anything, episode number six. We’ve got Rahul and our guest, Steve Brain with us today. And welcome, everyone. And it’s funny coincidence, we all happen to be in Seattle this time. Good morning, everyone. I don’t know if you have to adjust for time zones. Good morning. How’s everyone doing?

  4. Rahul

    Good morning. Yeah, it’s really strange. I mean, given that all of us are from different parts of the world…I just flew in from India. Steve came in from Utah, right, Steve?

  5. Steve

    Yeah. Yeah, from Park City. I kind of…bizarre weather. I left… It was snowing just last week there and it’s raining in Seattle, so summer’s escaping us in both places.

  6. Rahul

    Yeah. True. And yeah, it’s very rare for us to be in the same city though our views are slightly different. Stephen’s got LEGO in the background and I don’t this time. And Stephen, you had till very recently a view of the Space Needle right behind you.

  7. Steve

    Yeah. It’s just out there. I think the camera…oh, gosh. There you go, Space Needle.

  8. Rahul

    Yeah. And I’m in the suburbs here, so you get a variety of everything.

  9. Stephen

    Fantastic. It’s amazing how iconic that… I’ve got a friend who wants to come visit me because their favorite show was “Frasier” which aired 30 years ago but it’s still…it’s that anchoring image of that in the opener. So, we thought… Oh, and also before we start, I wanted to thank Ingrid and Christa, our ASL interpreters. They will be making the stream more accessible to everyone. And really appreciate what they do. It’s hard enough for me to keep up with this conversation in one language but to take it in one way and translate it in real-time, that’s a real skill. So, thank you so much for being here and making this more accessible to everyone.

    All right. Well, we wanted to start off this segment by going over some what’s new in AWS. So, I’ll put up some articles on the screen. Here we go. So, our first one that we wanted to talk about is Amazon AppStream 2.0 adds larger instance sizes to the general-purpose instance family. Oops. There we go. And I’m gonna put this URL in the comments so people can follow along if they’d like. So, what is this one? What’s the bulk of this announcement? Hey, Rahul, you had told me about this one.

  10. Rahul

    Yeah. So Amazon AppStream is actually, I feel, one of the rarely talked about services at AWS. They talk a lot about Lambda and Aurora and a whole bunch of other, you know, more glamorous services but AppStream, I feel, is one of the real heroes, your diamonds in the coal mine if you wanna call it that way. Because given that the vast majority of real-world workloads still reside in enterprises and on-prem, one of the easiest ways of delivering, you know, applications to your desktop users within a large enterprise is made so easy with AppStream because you can take it all in the cloud. And surprisingly, large number of these organizations are already familiar with the idea of hosting their enterprise applications as VMs and then streaming it via, you know, Citrix or VMware based products onto their desktops and this just makes it so much simpler because a lot of the other management overheads of your own clusters and stuff goes away.

    So, this is one of those solutions where…I think this came out almost seven, eight years ago, if I’m not mistaken. But you’ve rarely heard about it. You’ve rarely, you know, seen much discussion about it. But this is one where I feel large enterprises should really be looking at it because now there really isn’t any excuse for, you know, having entire teams of people in IT just managing and maintaining huge fleets of desktops where you have these desktop apps installed and you’re to keep patching them up, you’re to keep making sure that these are upgraded. Now everything gets delivered to you via the browser. So, I’m a big fan of AppStream because, again, taking workloads and moving them to the cloud made easy is something that it does brilliantly. And I just wish people would use more of it. Steve, any thoughts?

  11. Steve

    Yeah. I feel there’s a real applicability here in the remote world as well because it was one thing when enterprises had their Windows Boxes in the office and people would sit there and the IT guys could come patch the desktops. You would have the application data because now it’s a fat client in most of these cases, application data is on a desktop. When you look at a world that’s fully remote or partially remote post-COVID, the idea that you can actually host this in a secure environment and deliver that desktop experience to a remote device whether it’s owned by the enterprise or owned by the employee, there’s a lot of power there. I also think we… It’s really easy to underestimate the long tail of how long it will take to move from desktop to true kinda web-based cloud, and I was reminded of this coming up here. I was in the airport and we were having a question with a Delta flight and the Delta representative turned their screen around and showed us what they were doing. And they have this nice, modern web app and next to it was the green screen. It was actually blue in their case but it was a terminal application. And to do the job at check-in, they’re still doing a combination of web and a VT terminal. So, desktop apps [crosstalk 00:16:33] for decades.

  12. Rahul

    Yeah. Just one step up from there is to see the number of enterprise applications that still require you to install Internet Explorer and really older versions of them. You know, sometimes I have to figure out how do I get Internet Explorer installed to just access this application because I needed an outdated version of my OS which won’t allow me to do that anymore. And finding the right version of Internet Explorer to be, you know, compatible with whatever OS you have is just a nightmare. So yeah, I completely understand the long tail. It’s logistics forever.

  13. Steve

    Have you still got your Adobe Flash installation on a 3.5-inch floppy, Rahul?

  14. Rahul

    I not only have that but I have it in its original box.

  15. Stephen

    So, for those of our younger viewers, software did used to come in physical cardboard boxes and it was shrink-wrapped and you can go down to CompUSA and pick up things. I have a box copy of Caldera Linux somewhere that I bought from CompUSA.

  16. Rahul

    Okay. I can outdate you on that one, Stephen. I have 27 floppies of SCO UNIX where on the 26th floppy insert, you would have to enter something like a 32-character serial number and there is no backspace. There is no back arrow or something to go correct a mistake you made. If you made a mistake, you go back to the very first floppy and start all over. So, you know, I think we’ve come a long way and I’m so glad we have. I never want to install that one again.

  17. Steve

    I forgot about SCO. I can’t even remember who bought them. But it shows how the… Now Linux is what we’ll think of but, wow.

  18. Rahul

    Yeah.

  19. Stephen

    It’s funny because going back to the AppStream… I mean, not that long ago, I was in a company where I was one of their first remote workers and I remember going to DHL with bubble wrapping my laptop and mailing it to the IT department when I needed to because that was their process, was bring us the laptop and we’ll deal with it here. I said, “Okay. I’ll mail it to you and I’ll get a few days off.” So, it’s really great that we can now… I think it’ll take the burden off of IT departments to have not to manage and manage this physical infrastructure.

  20. Rahul

    Yeah. And with all these different instance sizes and choices that you have, like, I… You know, there’s very little that I see that can’t fit into this pattern of taking old desktop applications and putting them there. And I think Amazon is doing a pretty amazing job of, you know, giving so much choice and variety and flexibility. Like, now you don’t have to provision either too small or very large instances because, you know, the normal excuse that you see from consumers who move away from desktop is it’s not fast enough. And once you have larger instance sizes available, like, that’s a… even if you choose the largest instance size, it would invariably be cheaper than constantly upgrading your local machine and keeping it up to date. By doing it centrally, your overall costs are probably gonna be way cheaper. But you can provide the best experience with your applications with these super large instances. So, the larger you go, the better it gets with performance at least. And the fact that these are now available on AppStream is pretty awesome.

  21. Steve

    Yeah. And that’s a kind of indirect connection to something we’re talking about later with serverless which is… My daughter’s studying mechanical engineering and she needs to run CAD software on her laptop and she had a Mac laptop. The software is only available on Windows. And you need a super high-powered laptop to run CAD. But she’ll be running CAD software, you know, 2% of the time during an academic year. So, but what do you need to do? You need to go buy a Windows laptop to run the CAD software. And you think about that in the enterprise comparison which is now you’ve got people working remotely, bring your own device. You know, not only the security piece but what I really wanna do is pay for the consumption of these high-powered desktop apps when I need them, not all the time in a capital investment.

  22. Rahul

    Yeah. And I think on that…that’s a great point. A shout out to SOLIDWORKS which is, I think, one of the leading CAD software providers out there. Once when I first got into 3D printing, I was looking out for a tool to, you know, play around with and like you said, on a Mac, the number of CAD software that you could get, try out very quickly and decide which one you wanna go…it’s hard. But what I really appreciated about SOLIDWORKS was that they had… If you go and sign up today for a trial on SOLIDWORKS, they actually use AppStream to give you that full desktop environment available over a browser so you can sign up instantly, go try it out and you will see what CAD experience looks like with AppStream. So, if you can do CAD on AppStream, I don’t see why you can’t do most, you know, enterprise applications using it. It’s pretty remarkable.

  23. Stephen

    That’s really neat. So that way even if you’re browsing from your surface tablet and you decide that you wanna try SOLIDWORKS and you don’t really realize how resource intensive it will be and then you won’t be able to blame SOLIDWORKS later saying, “Hey, this is really slow.” No. It’s…okay, that’s great. That’s really neat. I’ll have to take another look at that. I’m still waiting for the Fusion 360 to fully support the Mac M1.

  24. Rahul

    Oh, yeah.

  25. Stephen

    All right. So should we switch stories and go to AWS Personalize? All right. So here’s the next one. Amazon Personalize adds support for unstructured text in six new languages. So unstructured, that means just free flowing, natural language, just…this is unstructured, what we’re doing right now. We’re just talking and there’s interesting words popping up, hopefully, somewhat often. So, and this is what this unstructured text is doing, right. It’s saying there’s text in product descriptions, reviews, movie synopses and this Amazon Personalize is taking this unstructured text and making it searchable and making it features into machine learning models…was that right? Have…?

  26. Steve

    This is correct. And it’s important to understand why it matters. So, take something…take eCommerce as a great example. You know, you go to the Amazon website or the Walmart website. What you see is a bunch of rich text descriptions, product descriptions, user-generated content like customer reviews and… But you also see some structured data, some schema data. You might have the weight of the product, or the type of the product, or the color of the product. So, what these companies are doing is adding structured data to their web pages to make it easier to search, to filter, to personalize, etc. But there’s two flaws in there, one of which is often the content that might be relevant to personalization isn’t exposed in that structured data. It might be content that’s relevant in a review like, where you live. Maybe the discussion in the review is about, “Hey, this is a great, you know, rain jacket for Seattle.” And the location’s not part of the structured data but the review’s now relevant to you.

    Another way of realizing this is each company that’s adding structured data to their website may be adding it in a different way or even across a website, different categories of products… You know, look on Amazon, different business…different retail units have different types of structured data on their products. So being able to spot unstructured means you can make better recommendations and recommendations that aren’t linked to what’s been externalized in that schema already.

  27. Stephen

    Are you thinking about the retail use case where there’s different dimensions that are important in different context, right? Like, you could be thinking about, say, for certain types of maybe backpacking equipment, the weight is what matters more than anything else. But then how do you…you’re not gonna have a certain… You’re not gonna have a JSON schema for every last clustering of products that you’ll have to be able to let a machine learning algorithm extract that for you seems a lot more powerful and a lot less manual work on the part of the retailer.

  28. Rahul

    Yeah, absolutely. For those of you who are curious about AWS Personalize or Amazon Personalize, we did a… We actually had a really great conversation with Ankur Mehrotra, the GM of AWS Personalize and Vertical AI. We’ll put out that podcast in the comments over here and probably even the video. But what I love about AWS Personalize is that they made personalization so simple. And one of the tradeoffs of making it really simple is that they have very discreet set of inputs that they require. The first one is you give it a list of users. The second one is you give it a list of your SKUs or, you know, the products that you have. And then the third one is list of interactions that your users have with the SKUs. And then the machine learning models basically decide how to send you your top 5 or top 10 recommendations based on that interaction data.

    Now that you have this description that it can automatically pick up from your SKUs and it can actually match on a bunch of other fuzzy parameters, actually makes personalization so much more powerful. I was seeing some of this stuff behind the scenes as you guys were talking about retail. I was reminded about…just last week before I was flying over to Seattle, I was looking for a microphone so that I could do this podcast from here or this live stream from here. And I was looking for a good podcast mic and I did not know which one to buy. And I can literally go and search for podcast mic for, you know, small room on Amazon and it brought up this Shure MV7. Like, I wouldn’t know to look for Shure MV7 or, you know, a very specific, you know, label or a tag that you would specify earlier. But everything that I needed is literally given there in the description of the product and they did a great job of providing a really awesome description, nailing down exactly the kind of customer who would benefit from a mic like this. And Amazon picked it up and it showed up as my top recommendation and I picked it up to, you know, buy it. So, I think this is incredibly powerful where business teams can focus on these defining really good descriptions about their products, nailing their customer fit and having Personalize do a great job of recommending appropriate items to their customers.

  29. Steve

    That’s a good, a really good example, Rahul, of why customer reviews matter as content there. Because in that case, it sounds like you…some of what you were looking for was in the product description but often, the application is what user-generated content delivers. A customer will say…like, you might go onto that website now and say, “Hey, I bought this Shure MV7 mic. It was an awesome podcast. I used it for my podcast, I used it in a small room. It was awesome.” And that just reinforces…it’s more information for the ML model as each review gets added.

  30. Rahul

    No, absolutely. Completely agree. I think that’s a great, great way to add a data source to your AWS Personalize. I had not thought about this. This is pretty awesome.

  31. Stephen

    I really like what you said about how the…what then matters is the description of what’s the actual submitted content. And it’s funny when you see… I suppose noncurated content like, on Facebook Marketplace or Craigslist. I was looking for a used car and you see this car that comes up that has nothing to do with your search. And in the description, “This car is definitely not a Cadillac. It’s not a Land Cruiser.” But what they’re doing is they’re adding keywords. Another one that I can think of is I saw… I was in Australia and I saw a Facebook Marketplace post for bacon seeds. I said, “What in the world is a bacon seed?” And it turns out someone had piglets that they were selling but in Australia, it’s illegal to transact livestock on Facebook Marketplace. And so, they were getting around that filter by saying bacon seeds.

    There’s always clever… And I think these…hopefully, then that will get added to the machine learning model and it’ll be some cat and mouse game of…but it’s been pretty fun. I thought that was actually quite clever on the part of the seller on how to use Facebook Marketplace in that way. All right. Let’s have a quick word from our sponsor and then switch to our last article for the day. So, one second here. All right.

  32. Ad Read

    Public cloud costs going up and your AWS bill growing without the right cost controls? CloudFix saves you 10% to 20% on your AWS bill by focusing on AWS recommended fixes that are 100% safe with 0 downtime and 0 degradation in performance. The best part? With your approval, CloudFix finds and implements AWS fixes to help you run more efficiently. Visit cloudfix.com for a free savings assessment.

  33. Stephen

    All right. And we’re back. So, let’s talk about this last one. This is the announcing AWS Cost Allocation Tag API. So, tags are a big part of AWS. You can take any resource and you can add key value pairs to it that can help you organize according to your own internal usage. But what are cost allocation tags and why are they different than regular tags?

  34. Rahul

    Steve, you wanna take this?

  35. Steve

    Yeah. So, cost allocation…so having tags is really important so that when you go to cost explorer, you look at cost and usage, you understand why you’re spending money. So as an operating CTO, and I’ve played that role at multiple companies, you’re often looking at your AWS bill and your CFO says, “Why are you spending all this money?” And you go into cost and usage and you’re like, “Oh, well, I’m spending 72% of it on EC2 and 18% of it on RDS.” But that’s not really actionable if I can break…unless I can break that down further. Now just looking at it by instance, now I’m in the weeds. And as I happen to know the IAE397263 is my, you know, directory service and then another instance is one of my web servers, probably don’t have that in my head and I need to aggregate it. So tagging is really important so that I can describe a class of service. Here’s my front-end web service. Here’s my directory service. Here’s my checkout service or whatever I’m describing it as, tags matter.

    The reason it matters to have an API is tags inhibit… Any company we acquire, often we’ll look and the tags just aren’t there because they’re hard. You have to go in the console each time you set an instance. So, you have to put them in the right place. You have to set the right tags. And you need to have a tagging policy. So, providing API access means I can now automate the application of tags and I can build it into my deployment pipeline and other things I’m doing. So, getting more and more complete coverage of tags is a real problem for any leader using AWS. And providing automation is…anything that’s built on humans doing something, is inherently always gonna have error and is inherently always gonna have gaps. So, automating…we’re engineers. Engineers use AWS. Engineering’s the solution, but engineering needs APIs.

  36. Rahul

    Another way that I describe the tags and this is… That AWS has a very opinionated view of the hierarchy or the way they see things. It’s their world view, their, you know, their visual lens. And that is very centered around regions, around their accounts, around organizations, around OUs, around services and specific resources. That’s their lens. That’s the way they slice and dice their data. And more often than not, it’s very, very different from the way businesses or customers of AWS see their world. The fact that your bill is coming up on a per-servers basis like Steve just said, makes no sense. Like, we have, what, 40,000 AWS accounts? And out of the 40,000 AWS accounts, if somebody told me, “Hey, you got an increase in your bill and it’s all because of EC2,” I wouldn’t even know where to get started. Like, I wouldn’t know what to do with it, right.

    So, the flexibility that AWS gives us is the concept of tags which allows you to define how you see the world and the fact that they allow you to tag practically anything you want. You can tag the resource, you can tag an account, you can tag, you know, specific kind of cost, you can tag a security group, you can tag a concept that you define of an application or something like that, a stack in cloud formation. You can tag pretty much anything you want. And you can then start operating along those tags as your dimensions rather than AWS’s world view. And I think that’s incredibly powerful. So, cost allocation tags came to the AWS world a few months ago, but the fact that they are now editable and you can update them and, you know, handle it kind of en masse with the API is incredibly valuable.

    Also, because this is post facto, a lot of customers already have hundreds of thousands of resources that are already deployed. Like, you are not going into the console, you know, going to every volume, going to every [inaudible 00:36:43] instance and saying, “Here’s the tag I want to add.” The fact that you can actually do all of that stuff now automatically and maintain the catalog via API is…and standardize it and make sure that it’s consistent… I think having the API is always incredibly powerful and valuable. By the way, my journey to AWS started when I first realized that you could deploy infrastructure with an API, that was my big aha moment for AWS and making the big bet on it because when you can do that, something as powerful as launching infrastructure…and I’m talking about back about 2006, 2007 timeframe where something like that was unheard of.

    Deploying infrastructure, you know, publicly in the cloud…you know, this was a time when we used to have people drive down to the data center because that data center was remote and that was a remote data center. You had somebody literally we would call in the middle of the night, have them, you know, drive up an hour or two to wherever the co-lo was, go into the rack, fiddle around with some wires and then be on the call where we were debugging to figure out what happened with the server. So, at that time when this API was available, it just was such a mind-blowing experience. And I think the fact that AWS started making sure that there’s an API for everything or nearly everything is so foundational to the way we think about and use AWS. So very, very valuable addition.

  37. Steve

    That is a… you know, and I was just thinking about that, Rahul. It really is a good reminder that the most important innovations in our world and in technology are in hindsight seen as obvious, yet at the time, far from it. And so, I was working…it’s interesting you said 2006. So, I worked at Amazon from 2005 to 2012. And when I joined Amazon, I joined a company where the employees themselves, particularly in retail, couldn’t really understand… They saw Jeff Bezos investing a lot of time and a lot of money in AWS, in video streaming, in Kindle, in Prime. And, you know, even the employees who were in the company, that had privileged information, they said, “Hey, this Prime thing makes no sense because it’s the most…the customers who shop the most use Prime, so it will never make money. Oh, video on demand. We don’t have enough bandwidth. It doesn’t make sense. The experience is really bad because of the bandwidth delivered to our houses. AWS? All you can do is store a blob and maybe spin up the Unix Box, Linux Box. What’s the value in that?”

  38. Stephen

    And aren’t we a bookstore? You know, [crosstalk 00:39:42].

  39. Steve

    Yeah. Yeah. We were a little bit beyond books by that time but yeah, there was a notion of, “Hey, we’re a retailer. We’re really good at that.” But the idea of, you know, some of these concepts, particularly AWS concept which was to go to the finest grain service and offer very simple capabilities via API and that was a… A guy called Rick Dalzell had a lot to do with that early on of just saying, “Take really simple services. Don’t try and offer everything.” So, it was very different to, say, the Microsoft approach of trying to bring the whole of the enterprise. They just started and they built, and they built, and they built. And we just spent time with AWS the last two days, the three of us and it’s clear they’re just, you know, continuing that very evolutionary approach of let’s just build another service and another set of APIs. And it’s fascinating to see how they’ve been able to continue to innovate by building in very much the same way. What capabilities can I offer behind an API? And it’s marginally incremental to customers.

  40. Rahul

    Yeah. I think that’s how I always think of the pace of innovation. You don’t know what you need to innovate over a period of time and I love Amazon’s approach of building these little pieces, putting it out there and seeing what sticks. But that allows them to be so much more agile, so much more nimble in how fast they can grow, how fast, you know, they can innovate on a whole lot of different dimensions. Not just, “Hey, we are gonna build us, you know, a message bus product or an eventing system.” And they spend all of their, you know, years trying to figure out what the most awesome, well integrated eventing system is ever going to be. But they had built out the entire eco…like, that’s…for me, that is monolithic, old-school slow-paced where instead…like, what they did with SQS. They started with SQS, what, 10…almost 10 years ago?

  41. Steve

    Oh, no, more than that.

  42. Rahul

    Yeah. More than that probably. Yeah.

  43. Steve

    We used to call SQS the queue that isn’t a queue.

  44. Rahul

    Yeah.

  45. Steve

    So, it was…yeah, it’s evolved so much. But, you know…

  46. Rahul

    Yeah. And then they figured out they needed EventBridge. They figured out they needed, you know, more sophisticated mechanisms of FIFO and stuff like that. So, you know, they learned and they evolved that ecosystem. But the fact that something like SQS was available as a super simple API call and you could pump in as much data as you wanted into it, as many messages…

  47. Stephen

    Went to beta in late 2004, GA, mid-2006. So, this is, Steve, right when you were starting.

  48. Rahul

    Well, yeah. Before that.

  49. Steve

    Yeah. And we…you know, again, inside Amazon, we were like, “This is a queue that’s not a queue. Why would we ever use this?” But, you know, what’s happened is it’s developed into this amazing service for serverless and event-based computing. SQS is a key enabler. There was a shareholder letter. This whole principle of doing many things, there was a shareholder letter from, I think, the early 2000s where Jeff Bezos and the Amazon executive teams, they talk about planting many trees and seeing what grows. And there’s some pretty famous failures. I mean, the Kindle Fire Phone was a tree that died, you know. It hardly got out of the ground. But [crosstalk 00:43:24] the number of successes Amazon have had really reinforces that model of decentralized investment, trying things, failing. There’s a lot we can learn from that.

  50. Rahul

    Absolutely.

  51. Stephen

    I tried to find the link to that discussion to post but apparently, Jeff Bezos is working on planting actual trees at the moment and that’s sort of [crosstalk 00:43:48] results. So [inaudible 00:43:53].

  52. Rahul

    You might want to look for let a 1,000 flowers bloom as the analogy. I’m wondering whether it was…because I’ve heard that one as well. So, I’m not sure if it was the trees or the flowers. It’s one of those two.

  53. Stephen

    [Crosstalk 00:44:09]. That’s pretty cool.

  54. Steve

    I’m pretty sure it was trees but that shows the problem of text, free text search. Sometimes you really want free text search with a good time-range algorithm versus all free text search.

  55. Stephen

    Absolutely. Well, let’s have another word from our sponsor and we’ll switch interpreters and we’ll switch gears into serverless when we get back. This was a great segment. Thanks.

  56. Ad Read

    Public cloud costs going up and your AWS bill growing without the right cost controls? CloudFix saves you 10% to 20% on your AWS bill by focusing on AWS recommended fixes that are 100% safe with 0 downtime and 0 degradation in performance. The best part? With your approval, CloudFix finds and implements AWS fixes to help you run more efficiently. Visit cloudfix.com for a free savings assessment.

  57. Stephen

    All right, awesome. And we’re back. So go ahead, Rahul.

  58. Rahul

    So, I’d like to just interrupt one second, Stephen. So, Steve, we kinda just…I just realized that we didn’t even have you give us an introduction at the beginning. We just got so excited to get started with, you know, all of this stuff. I mean, you put three techies in a call like this and that’s what’s bound to happen. But why don’t we just kick this off with you telling us a little bit about your background and your story for the audience? And yeah, then we’ll go back to Stephen.

  59. Steve

    Yeah. Yeah. So I’m Steve Brain. I run our technical product management group here at Trilogy, ASW Capital. And so, I work a lot with Rahul. My career basically has been three parts. I spent a bunch of time building high frequency and portfolio trading systems because pre the internet boom, that was actually one of the coolest places to do technology. Seems hard to believe now but trying to build fast moving trading systems, we were doing distributed computing and RPCs, three-tier, you know, early pre-internet. And then made the move to Amazon in 2005 and spent seven years at Amazon. And I worked on a bunch of really interesting things there and learned…I mean, accelerated my career. I learned so much at Amazon.

    And then in 2012, kind of lived the dream. My kids wanted to ski race so we moved. I had to leave Seattle and move to Park City, Utah and I moved into the startup phase. So, was an operating CTO for a number of tech companies. Qualtrics, did very well. InsideSales actually is now part of our portfolio. So, we bought the company that I formed their product. I left and then we ended up…they came back to me. So, I’m back involved in the same products which is a good, a really good experience because that’s a product we’re working on modernizing. It wasn’t on AWS but it was on some prior versions of AWS with ECS and now we’re looking at modernizing with serverless computing which is our topic for today. So that’s my intro.

  60. Stephen

    Oh, thank you for that. I realized the same. I was just so excited talking to you both and building on the momentum of the last few days. So, I guess for the audience, thank you, Steve. Really appreciate it. And I definitely wanna talk to you about some of the finance stuff. I started in finance, it seems like a lifetime ago too studying… I’m remembering studying market microstructure and some of the idiosyncrasies of that and how do you shave off a millisecond off of a moving average and all that kinda…that would be fun to talk about later. But now let’s get into what’s relevant for now, serverless. So, I guess as a first question that’s somewhat general or very general is what is serverless? So how would you talk to serverless for someone who is coming from a traditional, you know, like, a bank or a…some large institution that, you know, very well versed in technology but is used to the traditional monolithic application on the mainframe?

  61. Steve

    Yeah. The analogy I think of…if you’re coming from a bank, you own your own servers and you own the workload that runs on those servers. And that’s equivalent to owning a house, right. You own the house. It’s your house. You own it. Then what happens is you move to cloud and you can move it to a named server in the cloud. You can spin up an EC2 instance and you’re leasing that EC2 instance in the same way as you could lease an apartment. And, you know, you can move out of your house into an apartment. There are times that makes sense. There are times it, you know, might not. But you’ve now moved your workload. But it’s still in a particular address, a particular place. Serverless is the next step. It’s like, “Hey, I wanna come to Seattle for a few days.” So, I have a temporary requirement or a peak requirement. So, I rent a hotel room. And serverless is kind of the equivalent. It’s now, I’ve got this transaction. I’ve got this thing, this function, this activity, this right to a database and it’s moving to the point where I don’t manage the server, but I manage and pay for the level of the interaction on the transaction on the need.

  62. Stephen

    And then I like that analogy because you’re not going to, I don’t know, customize your hotel room, you’re not gonna put in a wall or anything but at the same time, you don’t have to deal with anything other than just being there. So, I really like that. That’s a great comparison.

  63. Steve

    Yep. I’m leaving today, I don’t even have to put the garbage out.

  64. Stephen

    All right. And so given… I guess we’ll switch to a more technical question. One of the serverless products from Amazon is Aurora, and that is a serverless database. And they’ve been innovating on that and they’ve released version two. What are your thoughts on the difference between Aurora serverless v1 and v2?

  65. Steve

    Yeah. I think what we’re seeing is them providing more granularity is how I think about serverless. Like, v1 let me introduce a database without having to commit to the server. I’m committing to the instance. And then provide granularity in which that can increase. What v2’s doing is giving me finer level granularity and more control on how I scale up and scale down. It’s still… When you compare it with other database products from Amazon, say, like Dynamo where you pay per write, per read, you know, it’s still quite…you’re paying for capacity on Aurora which makes sense with a relational database. But they’re introducing granularity. And what I would expect is, you know, there will be a v3 and a v4, however they name it, where they introduce more granularity. And it becomes easier and easier to scale up and scale down what I need.

  66. Rahul

    Yeah. Actually, part of me thinks, you know, every time I look at a new serverless service, you know, I think whoever coined the term serverless was a marketing genius. When you say, “Hey, I’m moving to the cloud,” you know, the kind of, you know…almost the kind of reactions I get span the spectrum of, “Oh, I love it,” to, “I have no clue what that is.” And there’s a really funny video, you know, on Indian television from about a decade and a half ago where one of the government officials is trying to explain what the cloud is and he’s literally confused with the cloud-cloud, okay. And it is an absolutely hilarious, hilarious video. But I think the idea of serverless at its core is that you’re making simpler, fewer tradeoffs. And literally to the name serverless, when you talk to somebody and say, “Look, you have servers now. What if you did not have the servers and you had the same service, right?”

    It’s such an easier…it’s so much easier to have that conversation than try to, A, explain to them the concept of the cloud, B, explain to them how they’re gonna get there. You know, it’s much easier to say, “You have a server today. Tomorrow, you don’t need the server and it’s still gonna be the same.” That’s a much, much easier conversation. So, whoever designed it or whoever coined the term serverless I think was an absolute genius from a communications and marketing standpoint. And back to Aurora serverless, yeah, I mean, back to the concept of tradeoffs that they were making. The first tradeoff they said was, “Hey, it’s hard to manage partitions and scale up and scale down and get it right every time. So why don’t you let us do that for you where we will just, you know, put all your servers in autoscaling group.”

    Kind of that’s the impression under the hood where you can just keep adding instances or removing instances and you get your, you know, horizontal scalability out of the box where AWS manages it for you. You could manage the same thing yourself by creating an autoscaling group and then, you know, adding that to the cluster and then creating your CloudWatch metrics to move it up or down. you could do that. But AWS has made it easier and said, “Hey, here is few things we will take care of for you. If you wanted to, set a max limit, set a min number of instances. We’ll do the rest for you.” That was great. That was the first version of tradeoffs that they reduced for us. And with serverless v2, they went a step further. They’re like, “All the increments can be a lot more granular, literally at the VCPU level.” So, they have something called the ACU or the Aurora Compute Unit.

    And they were like, “Why don’t we turn that down to a .5 ACU so you can move up or down incrementally in .5 units and we will make all the tradeoffs for you so you don’t have to worry about it? You would just put in your load and let it run.” Traditionally, how we used to manage it was we used to have to worry about the CPU load. We had a DBA…I mean, we had dedicated DBAs that would worry about CPU monitoring, memory monitoring, network IO, disk IO. God forbid one of the disks failed, you know, figure out a swap strategy. And then if you put in RAID, then the performance of the IO kinda went down. So, they had kind of some voodoo magic that they would do where they’d run RAID 0, but then have some mechanisms of doing backups so that they didn’t lose everything if something went wrong and then they had to work out all kinds of scaffolding for restoring, you know, all the data when things happened.

    AWS is basically saying, “Let us manage it and we’ll give you guarantees around it so you don’t have to deal with it and don’t build one-offs reinventing the wheel in every organization,” which I think is tremendous value for most organizations. And there may be a handful of organizations in the world that require that level of control for pieces of infrastructure like this, but for the vast majority, I think those tradeoffs are super easy to make because they’re so simple.

  67. Stephen

    Rahul, I’m really thinking about what you said about the genius of the serverless marketing. I’m trying to think. You know, would you call a car engineless? And okay. And what you really mean is you’re never gonna have to… I had a discussion…I remember when I was a teenager with this guy, a neighbor who worked at Microsoft and I said, “Would you want a car where the hood’s welded shut?” And he said, “Yes. I would love that. If they could sell one that’s so reliable that literally you couldn’t open it even if you wanted to.” And his response actually made me think that, “Okay. That’s a really good point.” You know, that was, you know, 20 years ago and now thinking about serverless, they’re saying, “We’ve got this covered so much that you don’t have to think about it. It’s not…don’t even think about any of the physicality, the backups, the power, the network, the bandwidth, nothing. You just think about what you’re putting in and what you’re getting out,” which is, like you said, it’s a very bold marketing naming but it works and it makes you think.

  68. Steve

    There’s some evolution in there as well, obviously. You know, you take the car analogy… I grew up in the U.K. where everything is a stick shift or manual as we call it. And coming to the U.S., right, almost all cars are automatic. And my late father used to say, “It’s automatic but you still have to drive it.” And it’s not really an automatic car, it’s just an automatic shift. But then you look at what Elon Musk is doing with Tesla. It’s becoming an automatic car, a car that can drive itself. Or it’s on the journey, we’re not fully there yet. So, I think if serverless in the same way is there’s many layers of serverless and we’ll see, you know, this become more and more a dream that you won’t be thinking even about how many VCPUs or, this thing’s there that some of the time it scales up and scales down but more just thinking about, like, my need versus the provisioning.

  69. Rahul

    Yep. And I think it’s always gonna be a balancing act as we, you know, see more of the serverless, you know, stuff come up. AWS, I mean, I don’t envy AWS product managers in this because they have to toe that fine line of how much control do they take away versus…you know, and manage it for customers versus leave control with the customers so that they can get more people adopting this in a, you know, far simpler, easier way. I know for a fact that, you know, serverless is now a big buzzword and a big initiative for all services at AWS. So, they’re going down this path. But for product management, that’s really the hard job of figuring out where you toe the line, what do you take on. Because when customers give up control, there are other tradeoffs that they may have to consider. So, yeah. It’s how do you make things easy and deliver a greater value where change doesn’t become your roadblock?

  70. Stephen

    Actually, that’s a really good segue of what I wanted to talk about next. This is a far more general question talking about all these different services. So, if we were starting out, how do we incorporate serverless into an existing application? It’s one thing to start from scratch and then whiteboard it and build this, you know, beautiful thing. But when you have an existing application or product that might be designed more traditionally, how do you bring serverless into the mix?

  71. Rahul

    Steve, you’re going to start with this one.

  72. Steve

    Yeah. So, it becomes a question of…you know, when you say an existing application, where do you start? So, we’re doing this right now. I mentioned InsideSales, there’s a product we own. It’s a product that was built about six, seven years ago. So, it was built as microservices, but we’re refactoring. You know, we had our own telephony stack and our telephony stack… The thing about telephony is you need to be running a process the whole time to listen for incoming calls, incoming SIP which is the control plane for telephony. So, you need to be listening on ports, you need to be ready to respond. That’s a product that we’re rebuilding using combination of APIs and serverless.

    So, we’re moving from our own telephony to using Twilio. When we interact with Twilio, we’re running Lambda functions. And how do we do that without throwing the product away is we found a clear seam that was an API where you could initiate a call, transfer a call, end a call, record a call. That seam was a place that we could cut along and without replacing the rest of the application, we could separate out hundreds of thousands of lines of telephony code and move it into this API-driven and serverless approach. So that’s typically what we do, you know. Sometimes you realize you have to replace everything, but often what you’re doing is refactoring behind seams in your application.

  73. Stephen

    Okay. So, you find there’s very clear places… Okay. This is one piece where you can almost excise out of the application, replace this with code and that… And right away there’s just couple of thousand or tens of thousands of lines that you’re no longer dealing with and managing and maintaining, but you’re handing it off to the AWS where they can do that for you.

  74. Steve

    Yeah. And I mean, another very simple example from the same domain is we had a problem when we rolled InsideSales out to Europe and we were dealing with European numbers and getting the phone numbers right, getting the country code. Sometimes somebody dialing, you get data submitted, right. People go in France, go on a website, give you their phone number. They don’t add the French country code. So now you need to get the right country codes if you’re running a call center, say, out of Ireland, a multilingual call center. So, we built a service that would take a phone number and would run it against Google’s libphonenumber library to see does this look like a valid phone number. And rather than make the sales rep wait when they press dial, it would state straight away you got the phone number wrong.

    So, simple value add service. I remember at the time we had a conversation, do we just embed libphonenumber? Do we put it inside of microservice? Now we’re taking that and moving it from a server-based microservice to a Lambda which is…it’s very simple. Here’s a phone number as an input. The response is valid or not valid. And one of the advantages in that case of putting it inside of a function is you can not only check libphonenumber, but once you get experience, we try to dial that number, it doesn’t connect. Let me recall that in Dynamo so that I actually make my algorithm as good as Google is. I’m getting real data and learning as I go. So factoring things correctly becomes a key point.

  75. Rahul

    Yeah. I go back to my default of where I usually start is literally using the term serverless. So, I look at all the servers that are there and I’m like, “That’s a database. Can we get rid of that database and just, you know, connect into an Aurora instance?” That’s the first serverless. “Oh, we have, you know, a cache which is a Redis cache. Can we do ElastiCache with, you know, the Redis instance type in it? That would be great.” Like, those kind of…there are just so many… You know, we have a queue, can we replace that with, you know, the managed Apache or, you know, Kafka queues or whatever, you know, service? Would that be a good replacement? Whether it’s Elasticsearch, can we replace that with OpenSearch, because we have to manage all these servers?

    That’s literally my first pass at taking a lot of the utility or commodity software that builds a lot of our applications and we spend inordinate amounts of time and effort managing and maintaining to see if that can literally be delegated to AWS to manage out of the box. And that way, thankfully, the seams are so beautifully defined and standard that you wouldn’t have to change anything in an application. You don’t have to go back to your development team and say, “Okay. I’m, you know, moving to this Aurora instance which is completely MySQL compatible or Postgres compatible.” You don’t have to worry about any of that stuff at all. It just works out of the box. And that’s my first pass.

    The second pass, absolutely what Steve was talking about where you’ll start looking at your application and seeing, “What did we custom build that could be replaced with something that’s commodity or utility, right? And that’s the second pass that you go through to carve out these pieces. And then the absolute third pass is the core code that you have written or what you think of as core. There are in the serverless world as you move from old monolithic architectures and stuff like that, there are just completely new and different ways of doing it which would also help you gut hundreds of thousands of lines of code and just leverage a framework or a mechanism out there instead of owning something of your own.

    So, the mental model at least that works for me is, I don’t believe that any product that was written over 10 to 15 years ago has more than 10% of the code base that they’ve written. That is truly business value driving. And the core IP, the core, you know, the thing that you wanna protect, it’s actually not even the code. I always keep talking about code being a liability. It’s never the code. It is what the code does. That’s about 10% of the millions of lines of code base, you know, that enterprises maintain and manage. The rest of it, you just have to keep adding, keep reducing and figuring out what can be carved out and have someone else manage it and not make it your headache.

  76. Steve

    One heuristic you can use here. You know, thinking about roles as one heuristic is that every time you have to do a root cause analysis of why my software failed, why a customer had an issue, why I have an outage, would be to say, “How do I make this somebody else’s problem? Yeah. My database logs ran out of disk space. Great. How do I not make that my problem? Hey, my box crashed or I had to do a security patch on the box or whatever the other reasons are. How do I make that not my problem?” And if you keep asking that question every time you look at a failure, what you should be left with eventually is that the value add, the tech diff of your application, the thing your application does that consumers pay you for. Because let’s be honest, customers don’t pay you for keeping your boxes up. They don’t pay you for keeping enough disk space for your logs. They pay you for whatever your product does that they can’t get somewhere else. And it’s really important to remember that. And all the other things we think about as engineers, hand that over to AWS or another cloud provider. You know, you’re not adding value. Nobody’s paying you to do that part of what your product does.

  77. Rahul

    Yeah. And I think the whole, you know, it’s not made here syndrome really needs to… I think organizations really need to think about that because if it’s not, like you said, core to your business, core to what value you’re adding, why would you own it? Why would you ever want to run it yourself? So, I think engineers have this sunken cost syndrome or the it’s not built here syndrome more than the business itself, the business side of the business. And I think we, as engineers, need to be more savvy to be able to recognize it, number one and two, to have a very pragmatic approach to say, “This is the value and therefore, we shouldn’t own it.

  78. Stephen

    It really is that shift in mindset, right, of… When you’re a younger engineer, it’s like, “Okay. Let’s just brew another cup of coffee and do an all-nighter. I can do this.” The phone validation is a perfect example where at first it sounds easy. Then you think, “Wait a minute. There are a thousand corner cases that you’re gonna have to deal with and what if someone needs an extension in their phone number or there’s countries that have variable amount of digits.” So how do you validate that? Or what happens if someone copy pastes in text in your text box that’s right to left instead of left to right? And how do you deal with that? Is the number backwards or is it forwards? And there’s a thousand corner cases that a team of professionals…that’s their fulltime job. It sounds a lot better than, you know, carrying this little hand-built library that…forever and maintaining that.

    Now unfortunately, it looks like we’re running low on time. This is such a fun conversation. I think, Steve, we’re gonna have to have you back again, if you don’t mind. I think there’s five or six questions queued up that I really wanna hear your thoughts about and I think the audience would as well. But unfortunately, we’ve run out of time for the day. So hopefully we can have you back in the next couple of weeks, if that works for you.

  79. Steve

    Absolutely. Would love to do it.

  80. Rahul

    Awesome. Thanks again, Steve. This was, as always, a fun conversation. And looking forward to the next one as well.

  81. Stephen

    All right. Thank you. And thank you, Christa and Ingrid for making this more accessible for our audience. All right. We’re gonna sign off for now and we will see you next time.

  82. Steve

    Thank you.

  83. Rahul

    Thank you.

  84. [01:11:17]

  85. [music]

  86. [01:11:38]

  87. Ad Read

    Is your AWS public cloud bill growing? While most solutions focus on visibility, CloudFix saves you 10% to 20% on your AWS bill by finding and implementing AWS recommended fixes that are 100% safe. There’s zero downtime and zero degradation in performance. We’ve helped organizations save millions of dollars across tens of thousands of AWS instances. Interested in seeing how much money you can save? Visit cloudfix.com to schedule a free assessment and see your yearly savings.