AWS Made Easy

Ask Us Anything: Episode 15

In this episode, Rahul and Stephen film from Anaheim, where they were attending an AWS Partner Summit. They filmed from a makeshift studio in a hotel room. Although the upload bandwidth could have been better, the topics still made for a great episode.

Latest podcast & videos


AWS announcement of private 5G

In this segment, Rahul and Stephen review the recent Jeff Barr blogpost on AWS Private 5G. The technology seems extremely interesting, but Rahul and Stephen have struggled to find a compelling use case. Most manufacturing facilities would already have wi-fi, and for new deployments, wi-fi seems like a more cost effective solution. However, they are open to learning more about situations where this is the best option.

cloud rating

NEW RDS Instances, Big discussion of stored procedures.

In this segment, Rahul and Stephen discuss some new instance types for RDS which have been deployed to some new regions. Although these instances are have loads of CPU power, the necessity for such instances means that your database is doing way too much computation. Through years of experience, Rahul and Stephen have found that increasingly complex business logic accumulates in database Stored Procedures, and it is the layering of these stored procs which put a big computational load on the database.

cloud rating

AWS Config has 20 new resource types. How does configuration drift happen? With AWS Config we can get alerts and automatic config fixing.

How complex is your deployment? Is it all in one place? Do you know every SAM deployment and CloudFormation template? If the answer is yes or no, you need configuration monitoring. With this announcement, AWS Config adds 20 new resource types. This is a building block for setting up configuration monitoring and alerts.

cloud rating

Amazon Personalize allows rule-based promotions in recommendations

In this segment, Rahul and Stephen discuss the new rule-based promotions in AWS Personalize. The main idea of this new feature is that, in addition to the ML-based recommendations which Personalize is famous for, there can now be rules for certain recommendations. For example, if you order a set of sockets, you may want to promote a set of extra 10mm sockets, since those are the ones most likely to be lost. The feature itself is really cool, but we would have liked a fun example in the article.

cloud rating

Amazon Rekognition Custom Labels now supports autoscaling of inference units

Amazon Rekognition is cool! Custom labels are cool! With this announcement, you can now autoscale inference units. This is a welcome change, but feels like the EC2 of a decade ago. Hopefully this is a small step on the way to fully serverless inference units.

cloud rating

Amazon Interactive Video Service

With this announcement, AWS Interactive Video Service now supports HD and full HD. This is a relatively straightforward announcement. We get additional features, with no increase in cost. IVS is a building block for interesting custom systems, such as transcribing a conference call. The videostream can be sent from the conference call to IVS, and then transcribed in real time, have sentiment analysis performed, and more. This is a straightforward article and announcement, and for this reason we award the article a “Simplifies” tag.

cloud rating


  1. Stephen

    Hello, welcome to AWS Made Easy, Ask Us Anything, episode 15. We are Stephen Barr and Rahul Subramaniam.

  2. Rahul

    Hey, everyone. As you can see, we have a new location and we’re in the same room, which is very unusual.

  3. Stephen

    The same continent.

  4. Rahul

    The same continent. In fact, we happen to be in Anaheim, California for the Summit. A lot of exciting stuff happening over here. So if you happen to visit the Summit here in Anaheim, please come to us, we’d love to catch up. And yeah, exciting talks planned.

  5. Stephen

    Two talks, very excited about those.

  6. Rahul


  7. Stephen

    I tweeted about those earlier. So make sure to go and look those up if you happen to be here, or just introduce yourself. And make sure if you are attending the event, to go to the [inaudible 00:10:31] and taco fix. That’s gonna be a fun after party to the event [inaudible 00:10:35].

  8. Rahul

    We have a lot of very, very fun stuff planned for the entire day tomorrow. So yeah, looking forward to meeting some of you who can join us over there. But we have an interesting lineup.

  9. Stephen

    Yeah. Well, first of all, how was your weekend other than, I guess, just packing and traveling?

  10. Rahul

    A lot of traveling, in fact. I didn’t realize that the flight all the way from India to L.A. was that long. I haven’t flown directly to L.A. So I flew by Dubai. And that was nearly a 16-hour flight. I think it’s been a while since I’ve done that long. So yeah, that’s interesting. And got through an entire book during the flight, so that is pretty good.

  11. Stephen

    I think you could do the entire “Lord of the Rings” extended versions. And if you were efficient, you could probably get “The Hobbit” there too.

  12. Rahul


  13. Stephen

    That’s a lot of flying.

  14. Rahul

    Yeah. So yeah, it was good. I actually got some good reading done on the flight. And, yeah, I’m just hoping that you don’t see me going away during the session. I’m still jet-lagged. I just came last night. But hopefully, we’ll get through this. There’s so many things we’re gonna talk about.

  15. Stephen

    Well, Seattle to Anaheim was a lot easier. I think it was under three hours.

  16. Rahul

    How was your weekend?

  17. Stephen

    Oh, very, very easy. I think we’ve had a lot of relatives coming in and out. So this weekend was mostly downtime with me and my wife and kids. And it was relaxing. We went to a little coastal community and just spent the day there. The kids [inaudible 00:12:08] and throwing rocks into the water and just having some relaxation. It was really nice.

  18. Rahul


  19. Stephen

    All right, well, let’s jump to our first segment which is a Jeff Barr article on the AWS Private 5G.

    Okay, so this is, Build Your Own Private Mobile Network. Okay, so [inaudible 00:12:41] so I wanna give a little bit of context. So mid-1990s, so I was, you know, eight to 10 years old. So I was one of these young families and five or six PCs in the basement. It was a really fun setup. And so my dad said, “Okay.” We bought a box containing a bunch of 3Com network cards, a hub, and drivers, and cables and set up a home network.

    He had this Vizio [SP] flowchart that had everything in there. I still remember. Because at that time, our ISP allocate us an entire class C. So we were all 209.20.218. It was really neat. Back then, they were giving them out. What are they gonna do with the addresses? These are the cards that we were using. I still remember this, the 3c509B combo card. And I remember going through the menus and setting every IR [inaudible 00:13:37] setting to make sure that my sound blaster and the network card work at the same time.

  20. Rahul

    Yeah, I remember these used WAX [SP] cables, right?

  21. Stephen

    Either that’s the combo card. You could use either the token ring or the bigger CAT [SP]. Well, at the time, it was a CAT2 but CAT5 cables.

  22. Rahul

    Yeah, I remember doing this coaxial and setting up [inaudible 00:14:00] network with these chips. Yeah, that’s pretty cool.

  23. Stephen

    So now fast forward a couple of decades and AWS is announcing private 5G networks. So this is an interesting one. Okay. We have…go to the blog post. Okay. It gives you a box of hardware and software. You can deploy your own private mobile network in a matter of days. So I guess when you sign up for this service, Amazon will give you the radio and the SIM cards. It’s 4G for now and 5G in the future, and AWS manages the infrastructure. You can get API’s for, I assume, registering a new SIM card or for changing some of the radio parameters. This is an area I know a lot about. So what’s your initial reaction to this?

  24. Rahul

    So I just found it funny that AWS 5G is really 4G. That was definitely interesting. But as I was getting through this, I was trying to understand where we were using this. So radio right now supports 100 SIMs, okay. And I think a lot of the use cases, the data out are about IoT in manufacturing use cases. But if you really think about it, you need an internet backbone somewhere in radio.

    So you’re gonna have the internet backbone anyway, and you have a manufacturing unit. Wouldn’t Wi-Fi just be easier? And you could use Wi-Fi as much as you want across the setup. Or if you had multiple, you know, locations spread, then you could use any of the ubiquity hardware or [inaudible 00:15:49] or whatever, to this point [inaudible 00:15:51] over a radio signal, and then have Wi-Fi locally.

    So I’m still struggling a little bit to understand who uses this, what, which customer. They haven’t put out a use case yet, which actually will really help to understand the use cases. I know that Amazon doesn’t build anything until it is a customer use case. So I would love to know who is actually using this and why. Also, when you look at pricing, this is pretty pricey, right? I mean, it’s about $10 an hour, if I’m not mistaken.

  25. Stephen

    Here we go. Pricing, each radio unit is $10 an hour with 60-day minimum. So we’re at 60 times 24 times 10.

  26. Rahul

    It’s almost 90k.

  27. Stephen

    Just to start.

  28. Rahul

    That’s about 90k a year to set up 100 radio devices, 100 [inaudible 00:16:43] points. That’s not much. With 90 grand, you could have state-of-the-art Wi-Fi all over the place. You could wire up every IoT device if you needed.

  29. Stephen

    So there’s got to be someone who reads this post and says this is exactly [inaudible 00:17:02]. We’ve been needing this, and Wi-Fi doesn’t work because of… Initially, I was envisioning this in some very isolated factory where they didn’t have cell service. But maybe it’s a place where it’s very crowded, or there’s some radio regulations around Wi-Fi, or the Wi-Fi spectrum is too crowded. I’m not sure. I’m trying to envision exactly [inaudible 00:17:26] someone out there has read this and said, “I need this.” But I’m jumping out to see who that is.

  30. Rahul

    Yeah, I think, I mean, that is one aspect to it, which is, this just seems really simple to set up. So if you are a provider who is trying to set up 5G, you know, something like a VNR [inaudible 00:17:50], this might be easier to get there. But yeah, I’m not entirely sure why or how someone uses it. It is really easy to set up. So you probably get that telco set up in a week with, you know, far less expense. And the time bucket is key, so you could probably do that. Right now, you have to piggyback on one of the, you know, service providers.

    So [inaudible 00:18:20] that if you wanted to just [inaudible 00:18:21] from scratch without [inaudible 00:18:23] you could just get a big pipe of internet bandwidth and provide your own services to whatever [inaudible 00:18:30] you want. But that every radio only supports 100 SIMs, that seems like a limitation [inaudible 00:18:38] of that and the price. I’m not sure.

  31. Stephen

    This definitely seems that you’d want to have predictable usage.

  32. Rahul

    Yeah, you would have a fixed number of, you know, IoT devices that are running on this. So again, I’d like to know what the real use case is or who the customer is that this is for.

  33. Stephen

    Well, this [inaudible 00:19:00]. At this point, they’re still adding the multiple radio unit features. So this might really be isolated for people out in a manufacturing facility or even a temporary construction job site something that [inaudible 00:19:14]. And maybe they don’t need an external. Maybe they just need to talk to each other in a way that 5G is better than Wi-Fi in this particular use case, maybe because of the device support or…

  34. Rahul

    Sure. Wi-Fi is across all devices [inaudible 00:19:29] today.

  35. Stephen

    Yeah, no, yeah. I’m not sure.

  36. Rahul

    I’m scratching my head. Yeah. So I think if I were to score this one, we’ll score it on one to five on content, right? I think I’d give this a two, simply because we still don’t know where and how to use it.

  37. Stephen

    Oh, I think that just for the fact, it would be really interesting if we know, okay, this is who is going to want to use this. That’s the most, I think, missing part of the… And maybe, I’m willing to bet that there’s someone out there who’s gonna read this and say, “Okay, this is the thing that solves my problem.” [inaudible 00:20:18] I just wanna build it. But it’s hard to know what it is. All right, so we’ll give that a two. Let’s do that.

  38. Rahul

    Yeah. The other thing that I would actually like is for them to have stated very clearly what the problem statement was. Right? When you look at every other AWS service, there’s a problem statement that that service tries to solve. Here, [inaudible 00:20:43] deploy 5G network. But why? Why do you need to deploy a 5G network, is something that I’m really confused. I know it takes a while. So is this meant for telco providers? Is it meant for [crosstalk 00:21:00]?

  39. Stephen

    Here we go. Okay. So on the product page itself, there’s smart manufacturing, security, campus connectivity, worker productivity. These are all…

  40. Rahul

    But Wi-Fi does all of this. Like, why is Wi-Fi not the right [inaudible 00:21:17]?

  41. Stephen

    Yeah. I guess that’s the question that this…the $90,000 a year question is, what is the advantage over Wi-Fi?

  42. Rahul

    Because at the end of the day, there’s an internet backbone that supports this. You have to plug in your [inaudible 00:21:35] and plug it into the [inaudible 00:21:39]. So all your communication is going through that network. So why 5G and not Wi-Fi? That’s what I’m really confused about.

  43. Stephen

    Yeah, all right. Oh, all right. Well, I guess we’ll look forward to learning more about this one. All right. Well, I think that’s…oh, and I guess for simplicity versus complexity, it’s almost too early to tell.

  44. Rahul

    Yeah, we don’t know the use case so it’s hard to say. We don’t have a use case to say, it makes life easier. Right now, I think given the [inaudible 00:22:17], the fact that we don’t understand it, [inaudible 00:22:23] a little bit.

  45. Stephen

    All right, so I have to come up with more graphics.

  46. Rahul

    Yeah. So I think leave it open, and we’ll try to get more clarity from the AWS team. And it’s good that there’s gonna be the Summit, so we’ll probably catch someone tomorrow or today, and see if we can get this question answered.

  47. Stephen

    Perfect. All right. Well, let’s switch gears over to RDS. Okay. This one is, Amazon RDS for SQL Server supports M6i, R6i, R5b instances in additional regions. So let’s see. First, the R-family is growing. So you did some research on trying to differentiate the R instances. What are your thoughts on them?

  48. Rahul

    Yeah. To be honest, it’s pretty confusing. So you have the R family, you have the R5, which are the old Intel processors running at 3.1 gigahertz. And you have the R5as, which are the AMD family of processors. And then you have the R5b, which are the latest generation Intel instances, also running the latest generation Intel [inaudible 00:23:51] that are running at 3.1 gigahertz but peak to 3.5 gigahertz. So that’s pretty interesting.

    But at the same time, you have the R6 family already there. So you have the R6as which are the latest EPYC processors for AMD. And then you have the R6g, which is the Graviton-based processors, right? And if you look at the price difference, the Intel ones, that is the R5 and the R6…I’m sorry, the R5b, are both about 20% more expensive than the AMD family. That’s the R6a as well as the R5a. I’m already getting very confused as I talk about this.

    But the R6g, which is the latest Graviton processor, is about 40% cheaper than the R5b, which is the latest one that Intel [crosstalk 00:24:49].

  49. Stephen

    So I think that our advice then is gonna always be the same, is stick with Graviton unless you have a really good reason not to.

  50. Rahul

    Correct. I mean, it makes absolute sense to just stick with the Graviton processors for RDS. And though, I’m not entirely sure if the SQL server specifically runs on the R6g.

  51. Stephen

    I see. So even though there are R6gs, that might be R6g for MySQL and Postgres.

  52. Rahul


  53. Stephen


  54. Rahul

    But there, the answer should basically be, I’ll have to check whether the SQL service will run on the Graviton processors. I doubt that they do. But at least the R6s, should run. There should be no reason to run the R5bs anymore. Because the R6as are already faster and, you know, cheaper, about 25% over the R5bs. So I don’t know why people would wanna use more R5s.

  55. Stephen

    If we look at the R6i itself, that’s a lot of compute in a database instance. Who are using these kind of…why do you need so much compute? I thought for databases, it’s mostly about memory and storage. Why need that much compute in a database?

  56. Rahul

    So the i instances are mostly about IOPS, okay? Now, [inaudible 00:26:01] need tons of IOPS [inaudible 00:26:04] memory. What happens to the database is you wanna take all of the stuff that’s in there, load as much as you possibly can into memory. And then, your [inaudible 00:26:13] operations are relatively straightforward because they just [inaudible 00:26:17] indexes in the data and pulling it out, right?

  57. Stephen

    Handling [crosstalk 00:26:22] and all that.

  58. Rahul

    So the CPU operation is relatively light when it comes to databases. It’s mostly about your disk IOPS to load stuff into memory, and how much memory [inaudible 00:26:31]. Now, the funny thing is that especially with SQL server, there’s a history where tons of products have leveraged, historically, [inaudible 00:26:47] procedures.

    So they’ve used these databases almost like compute, and I think that’s where the big problem lies. They stopped using the database like a database. They use it like an app server. There’s a lot of business logic kind of shoved into those databases. And I think that’s the wrong approach. I don’t think you should be doing that with databases. They were not originally designed to do that. There’s tons of hacks that have gone in.

    You want your database [inaudible 00:27:13] as light as possible in the relational model and keep the logic outside. Today, you don’t have any other tradeoffs to be made, whether it be at work or, you know, CPU or IO. You can get all of them out of the database.

    You know, back in the day, you had to move 10 [inaudible 00:27:30] out of DB into your app server. You would think twice. You know, you would really question how you’re gonna get that because the latencies were crazy.

  59. Stephen

    Yeah. So in this case, you always think they [inaudible 00:27:42] or I’m not limited. I just need to go to the 32x large, which is 128 CPUs.

  60. Rahul


  61. Stephen

    And then you’re not thinking, okay, am I…so you’re not bumping into this constraint where you’re thinking, am I really…is it really a good idea to put all of this business logic or data processing? I mean, I’ve seen machine-learning models built. Oh, there’s one place I did a consulting project and I saw a machine-learning model that was built out of a generated set of about 1400 stored procedures.

    And they were wondering, “Why is the database getting so slow?” And it was this incredible, just mountain of stored procedures that created thousands of intermediate tables [inaudible 00:28:25]. And it works in theory, but it was definitely not the right approach.

  62. Rahul

    Yeah. Well, that’s [inaudible 00:28:32] inability [inaudible 00:28:34] just untenable in the long run. I mean, I’ve seen products with [inaudible 00:28:38] and you don’t know how to deal with it. And migrating to newer versions becomes hard. Migrating off of, you know, the extensive licensing into more common databases is incredibly hard because the business logic is [inaudible 00:28:54] over there and you don’t know how to deal with it.

    So yeah, it’s a really tough space to be if you have tons of stored [inaudible 00:29:06] there is the more flexibility using the database itself for its core operation. If you look at the instances that we have, all of these instances have 12.5 Gbps at least, in terms of network bandwidth. Some of them have over 30 Gbps. If you look at R6i [inaudible 00:29:27], they use 37.

  63. Stephen


  64. Rahul

    Yeah, all the way up to 37. The R6i.metal goes up to 50.

  65. Stephen

    So you’ll be able to get your data out.

  66. Rahul

    You can get your data. I mean, the network is no longer constrained like it used to be, you know, even just a year or two ago. So yeah, stored procedures, I think it’s one of that bad practices that has come in with the relational developers and has stuck on.

  67. Stephen

    Well, and not just that, but often, they’re not implemented very well, they’re not… Yeah, it would be great if they had CI/CD around them but often, they’re just, you know, you open [inaudible 00:30:08] editor, write it down. And then someone else finds it and builds on it. And there’s not even a lot of thought that goes into actual [inaudible 00:30:15]. It’s just easy to find one view that already exists and build on it and another one, and you could have a lot of redundant computation. But [inaudible 00:30:26] really help with it.

  68. Rahul

    True. So you wanna get rid of as many of the stored [inaudible 00:30:32] as possible and move them into a business logic that you can manage, maintain. Just make your life easier. Keeping them in stored [inaudible 00:30:39] isn’t tenable in the long run.

  69. Stephen

    Yeah, absolutely. So coming back to this article itself, RDS for SQL server. What do we…I mean, this obviously is simplicity in the sense that if you [inaudible 00:30:56] that they do, life is better.

  70. Rahul

    Yeah. I think so M6i, R6i completely make sense. Definitely [inaudible 00:31:05]. The one point I would ding for is the R5b because I don’t understand why you would want to launch the R5b. The R6a should be a better option for a SQL server.

  71. Stephen

    Is it… I guess, so they’re saying R5b compared to R5. But then why wouldn’t you just go to R6?

  72. Rahul

    Yeah. R6a is already there. R6a was launched some time ago. And R6a is 20% cheaper than the R5b. It’s already the next-generation EPYC processors. It is [inaudible 00:31:39]. It’s just not Intel. And your RDS SQL server should run just fine on that. So why wouldn’t you use R6a against the R5b?

  73. Stephen

    That would be a really interesting bit if they said, “Okay, the people who are going to be using the R5bs, what would…you know, in the big picture, what’s stopping you from switching to an R6 instance?” So…

  74. Rahul

    Yeah. There’s nothing.

  75. Stephen

    So what do we put in terms of our rating today? I’m thinking about 3.5. I mean, it’s very clear there’s nothing wrong, it’s got examples, it’s got all the data you need. It would be nice if there was a bit more bigger picture. I mean, that’s not really…yeah.

  76. Rahul

    Yeah, I’d give it 3.5. Do we have a half? No, [crosstalk 00:32:32].

  77. Stephen

    There’s a little half cloud there.

  78. Rahul

    So you give it 3.5 clouds. Okay.

  79. Stephen

    Let’s take a quick break. And then when we get back, we are going to be talking… What is our next one? Oh, we’re gonna be talking about the interactive video service. We’ll be right back.

  80. Woman

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

  81. Stephen

    All right, this segment, we’re talking with Amazon interactive video service. Okay, so this is…what is interactive video service? This is a live streaming platform. Okay, we can use interactive service basic channel for HD 720p full HD. And then this is expanded functionality, designed enable streamers and viewers globally to enjoy higher video quality when using a basic channel.

    Okay, so who are the users of interact video service?

  82. Rahul

    So the interactive video service is really cool. Think of the interactive service as taking a live stream, and then adding or embedding that with a larger application that can do interesting stuff with that live stream. So, you know, for example, if you were doing a sports broadcast, you have your commentary and your live broadcast going on. But at the same time, imagine being able to present all the stats or predictions about what’s going to happen in the game, doing all of that live, as the video stream is coming. That’s incredible. That’s basically what the IVS service is all about.

    We use IVS. In fact, we have to demo this, and it works better when there are multiple [crosstalk 00:34:38]. So once I’m back in India and broadcasting from there, we’ll definitely do a demo session. I think that’s a pretty cool. So what we have done with IVS is we have a lot of meetings that we do in the organization, you know, when you have a few thousand people [crosstalk 00:34:57].

  83. Stephen

    That are remote.

  84. Rahul

    Yeah, completely remote. These [inaudible 00:35:01] are a big part of our organizational knowledge and organization knowledge base. And one of the challenges of remote, as we all know, is when we’re in these Zoom calls or a Chime meeting or whatever, ends up monologue, people become disengaged. And the speakers usually have no idea about engagement in the audience. You know, you can have somebody drifting away and doing other stuff.

    So we’ve created a little IVS, you know, solution, which takes our Zoom and Chime meetings and so on, and it does real-time live analysis of what’s going on in the meeting.

  85. Stephen

    Building on Amazon’s other products.

  86. Rahul

    Correct. It uses Comprehend to understand what the conversation is all about, so you can search for it later, if it’s recorded. You can search for it. You can, you know, figure what the meeting is about, and you go back to it. You can refer to it based on the keywords that [inaudible 00:35:58] finds in the topics [inaudible 00:35:59] finds.

    Of course, you can train it with custom vocabulary like [inaudible 00:36:04]. So you can do all that stuff with it. The other thing that it does is it also uses video stream recognition. So it looks at, you know, facial sentiment, so whether everybody’s confused, disengaged, happy, or angry, or agitated. We wanna make sure that we are…you know, that the feedback comes back to the speaker. And, you know, generally, as part of the [inaudible 00:36:31], understand the audience and what kind of a mood.

  87. Stephen

    It’ll be interesting to have all that data and you loaded it into a graph database. You could do queries. Like, “Tell me if anybody who becomes confused after a certain person speaks. And then what do they do? What do they speak about afterward? Or how’s their engagement in other meetings throughout the day.”

    And it’s not about surveillance. It’s more about trying to make sure you’re effective, right? Because when you’re in in-person meetings, you get all these verbal cues. You get to see people, I don’t know, just reaching checking their cell phone. But Zoom, you only see what they want you to see, which is great. But also, it’s hard to then gauge your audience. Just like right now, I don’t know how they’re responding to this.

    And you just have to get used to staring into the camera and hoping it works. But getting feedback for engagement is really useful. And, you know, overall, I think, you know, [inaudible 00:37:28] benefits of remote would outweigh the costs. But one thing we have to [inaudible 00:37:35] is, you know, try to get better engagement in meetings.

  88. Rahul

    Great. And engagement is the biggest challenge of remote work. So, you know, one of the other things that’s really hard to figure out, we rely so much on body language, that kind of feedback. You just don’t get that in Zoom, or a Chime meeting or, you know, whatever mechanism you’re using for remote operation.

    So we are trying to use IVS and AWS information and a bunch of other tools behind the scenes, to figure out, you know, whether people are engaged or not. By the way, another kind of conversation that is not very engaging is the one where you have one literally in the monologue. If you have 10 people who are just literally staring at a screen, that’s not an engaging conversation. That’s not the kind of conversation you want to encourage within an organization.

    So you want, you know, everyone to interact. That’s the whole point of these meetings. If it’s a monologue, someone should just record it and post it. You shouldn’t be having a meeting wasting everyone’s time on it.

  89. Stephen

    I saw a blog post, it was on Hacker News a few weeks ago, where a guy built himself a bot because he was a person who recognized he liked to ramble. And so it started warning him if he had gone on more than a two-minute tangent, something he’d built personally for himself, a little Zoom bot. But this is something that you could use IVS for to look at even just the distribution of speakers throughout a conversation.

    Because and then also, if you have people who aren’t contributing to meetings, and it’s a good tool to know, okay, why aren’t they is there? Is there something about the leadership style? Or is it something that can be changed to increase the engagement of all the participants?

  90. Rahul

    Yeah, it also has an application in all of our DEI [SP] efforts. Like, inclusion is a big part of what we’re trying to do. And when you’re in meetings and only one or two people are interacting, you know, as a speaker, you’re so focused on weighing your view and your vision or whatever else and you miss out on the fact that one’s interacting. And if you’re, you know, IVS dashboard shows you that, “Hey, this person hasn’t interacted, you should engage proactively with person A and C.” And as per your communication, your messaging, you can actually ask that person the question and say, “Do you have any questions? Do you agree with this or not agree with it?” And it really helps with making sure that the conversation is a lot more inclusive.

    And it separates out two modes of communication. One is a meeting where you want everyone to collaborate and work together. And the other one is a broadcast where you should record your views and put them up, and people watch them asynchronously and comment on it or whatever. By the way, we have an interesting solution on that also, we can probably bring that up during one of the sessions.

    So we do these little snippets of videos we record, and we allow you to review them [inaudible 00:40:39], recommend them. And that’s really easy to communicate in a more natural manner than having to write these long documents. So yeah, it’s a very interesting platform. So the fact that now it can support in full HD, that’s pretty awesome.

  91. Stephen

    That’s funny. I was just thinking about that in the context of, you know, old hardware. We brought up those 3Com, 3C [inaudible 00:41:02] earlier. And I remember, you know, paired with that was my little quick cam, you know, going into the parallel port. I think it was 60P. I think [inaudible 00:41:11] to this. It did, what, 320 by 240. And so, now thinking that you can play with these 10 ADP streams, it’s pretty fantastic that you could have that as just another named object that you can then move around and analyze and have the bandwidth to be able to do that. We’re streaming right now from a hotel, and it seems to be fine.

  92. Rahul

    Yeah, by the way, the oldest video conferencing equipment that I ever had was this one megapixel. And I don’t think we even had a P associated with it. I’d have to work out what that would be. But it would typically…it was just a one megapixel [inaudible 00:41:52] device that sat on top of a CRT television, and you connected it to your CRT television for video out, over those analog cables.

  93. Stephen

    And it probably cost $1,000.

  94. Rahul

    Yeah, it was pretty expensive. And it had those analog cables that would connect in to your CRT televisions.

  95. Stephen

    Okay, the yellow and red and white ones with the RCA ones. Okay.

  96. Rahul

    Exactly, [inaudible 00:42:13] cables. And then you had your ethernet into it. They would connect in to your internet connection. And then your only connect point, IP to IP.

  97. Stephen

    I can picture it, okay.

  98. Rahul

    So picture it and that stuff, you know, was simple enough for you to navigate through it or set up port forwarding and solve all of the issues. So that you actually connect point to point between those two [inaudible 00:42:40].

  99. Stephen

    Oh, it had to do with the actual router itself and…

  100. Rahul

    You had to manage all of that. It had no mechanism for doing any of that stuff. And from that, we moved to [inaudible 00:42:51] by Logitech I think at one point in time. I think over the years, we started remote work back in 2010. 2008, the conferencing was in its infancy. Nothing like what we have today. So I’m glad that we have full HD [inaudible 00:43:15]. If we can do [inaudible 00:43:16]. Are we broadcasting in 4k here?

  101. Stephen

    I think we are. I’ll have to check the stream. Yeah, we’re broadcasting whatever is supported by StreamYard is what we’re doing.

  102. Rahul

    Yeah, and we are sitting in a hotel room, in a makeshift, you know, setup.

  103. Stephen

    We gotta get a picture of the studio before we take it down. This is pretty good.

  104. Rahul


  105. Stephen

    All right. So interactive video service. How do we rate this? I think…

  106. Rahul

    This one is relatively simple and straightforward. So I think in terms of content, it definitely simplifies. In fact, more than simplifies it, it actually makes the service more useful in the current context. And yeah, I would rate this a four or five.

  107. Stephen

    All right.

  108. Rahul

    Yeah, this is pretty straightforward and simple.

  109. Stephen

    Yeah, nothing…yeah, it’s a new feature. Very much that…yeah, how could you complain having now you can do 720p? It’s pretty good. All right. Well, let’s go over to Personalize. Oh, no, no. We are doing AWS Config. All right. Here we go.

    All right. AWS Config supports 20 new source types. AWS Config, I was thinking about, this is an idea that it’s configuration monitoring. But I was thinking about how complex configurations have gotten. It used to be you’d have, like I mentioned earlier, one [inaudible 00:44:50] flow chart that’s everything. That’s the whole network.

    But now, it’s almost there’s urgent behavior, right, and you have different resources that are coming and going in a very ephemeral nature. You have different scripts that people have written who are monitoring and making changes. And overall, you reach the same state, but then you want to make sure that you don’t have any of these runaway behaviors. And that’s where Config can really help you.

    You can monitor from a big picture and you can monitor on a per resource type. And so what this is saying, this article, there’s new supported entities for Config. Here’s the [inaudible 00:45:31] right here.

  110. Rahul

    Yeah. I think Config monitoring and management is insanely complex. I mean, take a look at the first one, the routing table, itself. So let’s say you have an organization giving its course.

  111. Stephen

    Oh, that’s the previous one?

  112. Rahul

    Yeah, that’s the previous one.

  113. Stephen

    Oh, okay.

  114. Rahul

    So take a look at the routing tables, for example. You know, we have about 30,000 AWS accounts and with a very, very complex routing scheme across all of those different VCs and all of those different accounts. And these routing tables get pretty darn long.

  115. Stephen


  116. Rahul

    And you [inaudible 00:46:14] that there isn’t any drift that happens in configurations. Let’s say there are some servers that [inaudible 00:46:21] TransitGateway. You don’t want that unless it’s authorized, right? You want the table to open that Config changes. You want the version, you wanna set Config for a particular server, a particular device, for a particular [inaudible 00:46:34]. You wanna make sure that the changes that are happening in the configuration are controlled and you don’t have random services and API making those changes. You want to detect them, you want to act on it.

    So I think having more and more support for detecting these kind of configuration changes is incredibly valuable. You want to get an alert whenever it happens, and you act on it. In most cases, you are going to basically revert. So for any organization that is centrally managed, you want to be able to revert any change that happens as it happens [inaudible 00:47:11] configuration management system.

  117. Stephen


  118. Rahul

    So you want to immediately, even if someone has admin permissions to a local [crosstalk 00:47:19].

  119. Stephen

    So how does it happen? If someone, let’s say, ran a script and they ran a script with the wrong environment variable. And all of a sudden, what they thought was their test account is not the permission account.

  120. Rahul

    I think it’s a little more basic than that. You might…somebody [inaudible 00:47:37] operating and it’s too restrictive. So because they’re an admin, have access to that particular account, they go ahead and open up a port.

  121. Stephen

    Open up a port?

  122. Rahul

    Or they open up the network to other traffic that’s coming in, just to make their lives easier. Like, I’ve seen so many people [inaudible 00:47:56] their security groups [inaudible 00:47:59]. And that’s a terrible idea. You don’t want people to create, you know, [inaudible 00:48:08]. And override it locally. So at the AWS organization level, you have a configuration already set. Anytime you make a change in the Config, you get an alert. And then act on it and write something back.

  123. Stephen

    Yeah, that makes a lot of sense. Because as much as you’d want only one single person or single entity that’s managing the configuration, that’s just not practical for a large organization. But to be able to see these changes and say, wait, this is a security violation or a cost overrun, it’s going to happen if you direct this. Like you said earlier about limiting that blast radius, right? You wanna have things in place, mechanisms that will eliminate or contain your blast rate. Or if something goes…something’s misconfigured, you don’t want to find out after the billing cycle. You want to find out right away.

  124. Rahul

    Exactly. And if we look at a set of these services, right, the other mechanism which configures use a lot is version control. You wanna make sure that you don’t haphazardly update or hold back versions of configuration across all of these different systems. Whether it be your Notebooks, whether it be your data club, you wanna keep control over all the versions. And in this Config app, you manage all of that.

    Anytime you see that change, you get an alert, and you can go ahead and roll back or manage the version accordingly, okay? You want all of these…when you want better control over people randomly changing stuff, Config is the service you go to, to make sure that things are not kind of happening very haphazardly.

  125. Stephen

    How much…what percentage of AWS entity do you think our [inaudible 00:50:02] Config?

  126. Rahul

    Right now, look at all the different places and all the different points. I would say just short of 50 is my guess. I don’t have the full map, but I’d just say just short of 50%.

  127. Stephen

    Given the speed at which they’re rolling out new things, that’s pretty good.

  128. Rahul

    Correct. Yeah, I think it’s pretty good. I mean, there are a bunch of services that are always playing catch up to the main service deployments.

  129. Stephen

    And this is going to be one of them.

  130. Rahul

    This is gonna be one of them, right. [inaudible 00:50:34] endpoints is the one where, you know, this comes out first, then you come out with the [inaudible 00:50:38] endpoints. You wanna roll out the basic services first, and all the other [inaudible 00:50:43]. Like, [inaudible 00:50:46] from the standpoint across all of these different, you know, services. It just takes them time to get those things plugged in.

    So yeah, I’d say the Config always plays catchup. But I think they’ve made good progress.

  131. Stephen

    So then, thinking about this article, I guess in our skew, what do we say is our rating for this one?

  132. Rahul

    So I would say this is a good 4.5 to 5.0 because it’s pretty clear and obvious.

  133. Stephen

    Wow, that’s pretty…four. Yeah, and I think also the simplified. Because it’s just 20 new resources, it tells you exactly what they are. If you’re a Config user, okay, great. I’m using TransitGateway routing tables and now I wanna bring those into my Config. And set that up that you can [inaudible 00:51:41] those.

  134. Rahul

    Right. The traditional way of doing this was that you would have an event where something went down, or some configuration is a problem. And then, you would say, “Let’s go look [inaudible 00:51:54] to figure out who accessed this API to make changes to this particular service.” And then you will dig. And then you will go back to the person, and then you would know, you know, what config to bring back. And it would be [inaudible 00:52:06].

    So [inaudible 00:52:07] in real time.

  135. Stephen

    So I mean, that stuff is still good to know, but it’s good to know after you’ve reverted your changes.

  136. Rahul

    Exactly. So this gives you much tighter control for that kind of scenario.

  137. Stephen

    All right. Well, let’s take another 30-second break and when we come back, we’re gonna be talking about personalize and recognition.

  138. Woman

    Is your AWS bill going up? Cloud Fix makes AWS cost savings easy and helps you with your cloud hygiene. Think of Cloud Fix as the Norton Utilities for AWS. AWS recommends hundreds of new fixes each year to help you run more efficiently and save money. But it’s difficult to track all these recommendations and tedious to implement them. Stay on top of AWS recommended savings opportunities, and continuously save 25% off your AWS bill. Visit for a free savings assessment.

  139. Stephen

    All right. Amazon Personalize allows rule-based promotions in recommendations. This is pretty neat and it’s pretty straightforward, right? So the idea is you have a recommendation engine, it’s sitting there. I don’t know, you bought a sandwich maker, maybe you need a spatula or…I don’t know. Something…

  140. Rahul

    They recommend to you a spatula or whatever, depending on what you bought. You know, typically, when you buy something, it says, “Other people who bought this also bought these particular items.” So depending on the parameters that are provided, the recommendation engine will give you a bunch of recommendations. You decide how many you want.

  141. Stephen

    Yeah, there’s some really bad ones. Have you seen where you buy a ski mask and it recommends a baseball bat?

  142. Rahul

    Yes, yes, I’ve had some bad recommendations [inaudible 00:54:00] just look at the Amazon Store. Based on the stuff I usually buy, the recommendations are almost spot on. Like, especially when I’m buying groceries and stuff like that. The more data that the system has about my interaction with the Amazon shopping carts and so on, the better the predictions get.

  143. Stephen


  144. Rahul

    Like, once it realizes that I’m ordering something every month, it says…it recommends that make it a monthly order, rather than make it, you know, an ad hoc, you know, one-time thing. And it recognizes all of those patterns pretty intelligently. So AWS Personalize is really awesome in that sense, from that standpoint. And I think the very first episode we did, the other podcast, AWS Insiders…by the way, that you can find on as well. We have our very first…

  145. Stephen

    Oh, here we go. It’s on Apple Podcasts. I’ll put it on

  146. Rahul

    Yeah, it is on as well. So find that podcast there where we talked to Ankur Mehrotra, the GM of AWS Personalize. There you go. So go to that one. And the very first episode, we talk about AWS Personalize. It’s really very cool. A lot of very interesting scenarios.

    What I like about this particular announcement is that [inaudible 00:55:30] things to do is when you have a set of recommendations, you want the recommendations to, A, be stuff that the engine finds. But somewhere in there, you also want to recommend things that you want to push like promotions. Right? These are things that you want to push to all users irrespective of what the recommendation sets are.

    And that has historically been hard because you get a recommendation set and you have to figure out which one of those five or six items you want to eliminate to insert one of your promotions. And then how do you figure out the distribution? Do you promote the same stuff to all of your users? Or do you promote it only to a set of those users? So this system really helps you solve that problem where you can say, “I want 20% of the recommendations to have all the [inaudible 00:56:21] items to go to all of these users.”

  147. Stephen

    Yeah, okay.

  148. Rahul

    And so it’ll embed those inside recommendations based on also what is relevant to them. So you can say 20% of the recommendations should be stuff that I want to push, right? And the AWS Personalize will take care of that for you. It’ll still seem like all very relevant recommendations and the engine takes care of all the complexity for you.

  149. Stephen

    This is gonna be really useful for retailers who have particular things that they want to sell and know certain combinations go really well together, certain pairs of products. A long time ago, not that long but when I was buying baby things on Amazon, it was clustered. There’s people who like Disney branded stuff, or there’s people who like organic, natural stuff, and just kind of clusters of [inaudible 00:57:16] in the baby gear world.

    I can see being able to insert a new product into the mix. And instead of waiting for the recommendation system to pick it up, to be able to promote it. That’s a really neat idea.

  150. Rahul

    I mean, one example, I think I read a case study some time ago, where if you are buying baby products on Amazon, they always have a diaper in the recommendation. As far as I know, Amazon has never made a profit on the [inaudible 00:57:48], if I’m not mistaken, at one point in time. But it’s just such an essential part of the shopping experience. Like, if you don’t shop for diapers on Amazon, you would probably not buy the other products over there. So…

  151. Stephen

    Yeah, and you’d never ever see that and go, “I don’t need diapers.”

  152. Rahul

    Exactly. So they would look for it, and this is a mechanism that allows you to do that. And it’s not just retail. I think there’s a lot of application in any kind of content that you’re serving out to individuals, whether it be news, whether it be movies, whether it be music. In all of those scenarios, you have a situation where there’s a new show and you want to recommend a show.

    Like, if we were to have a recommendation engine for someone visiting our AWS Made Easy site, we’d have to do a top five. Then in today’s show, we would push AWS Personalize and all the content that we have around Personalize. Because that’s what we’re talking about. Right? That’s something new, that’s a new announcement. So you can push that content.

  153. Stephen

    And that’s the other example it gives, right? It says, “Video-on-Demand customers can use 40% of their rail/carousel with linear live TV content.”

  154. Rahul

    Right. Or any new releases, like a new movie just released or [inaudible 00:58:59]. They always have it part of the recommendations. So yeah, I mean, this makes life easy for anyone who is doing tons of recommendations as part of their application without having to manage the complexity of how to present all the data and inject these recommendations in the middle of other standard, you know, [crosstalk 00:59:24].

  155. Stephen

    Yeah, because otherwise, you’d have to write some code after Personalize to then say, okay, take this out, put this in. And then you’d have to watch it and monitor it somewhere else. So this brings it all under one place. That’s a really neat idea.

  156. Rahul


  157. Stephen

    All right. Well, I guess in rating, I’m gonna give this a simplifies. And in terms of the article content, it would have been neat to have a hard example. I mean, they give soft examples in terms of this is the kind of thing you can do. But it would have been neat to have, I don’t know, a concrete example to go on.

  158. Rahul

    So I think would I really love for something like this is, like, the [inaudible 01:00:04] article that we reviewed in the beginning where there’s a step-by-step process of how you set this up. Having that step by step, you know, here’s how you do it in the console, here’s how you set up a new rule, that would have been really neat. Of course, this is fairly intuitive. You could probably just go to the console and figure it out. But from a content standpoint, I think having that would have been really useful.

  159. Stephen

    Yeah. Well, all right, so let’s give that a…what do you think? Three and a half?

  160. Rahul

    I think it’s a 3.5.

  161. Stephen

    All right.

  162. Rahul

    Yeah, 3.5 seems good.

  163. Stephen

    There we good. Oh, there’s a little half cloud there. Let’s see. We’ll have to get some outlines in these graphics.

  164. Rahul

    We’re gonna make the clouds grey, I think. Or put in an Amazon, you know, a yellow or orange, whatever the shade is.

  165. Stephen


  166. Rahul

    The orange border around the cloud.

  167. Stephen

    We can do that. Okay. Well, let’s go to our next one, which is Rekognition.

    All right. Amazon Rekognition custom labels now supports autoscaling for inference units. I think my initial reaction to this is, okay, that’s great, but IUs are really expensive. So this is interesting. So in and of itself, custom labels, the idea is you have a video stream or an image stream, and you want to look for particular objects that are gonna be in the standard training set, in the standard catalog.

  168. Rahul

    In the standard catalog.

  169. Stephen

    Maybe in a manufacturing scenario or, I don’t know, a security scenario where you’re doing something different, and you want to detect these and add these custom labels. So okay, so previously, Custom Labels with unpredictable workloads had to set minimum inference units to support the peak volume. So this seems very…this sounds like 10 years ago, right? Oh, we don’t know how big of a load you’re gonna need, so just provision for the maximum load. But that sounds very old.

  170. Rahul

    This is the on-premise thinking for a cloud service.

  171. Stephen


  172. Rahul

    But I’m glad that they’re changing that.

  173. Stephen

    So now, you can. So but I guess what surprises me about these is that, okay, now you’re only charged $176 for 19 hours plus 5 hours, 24. 24. So the use case they’re giving is that you had to scale up for five hours out of a 24-hour day and that only costs you $176 instead of $480. This still seems really expensive, I guess is my reaction to this, in general.

  174. Rahul

    Yeah. I think so right now, for the inference [inaudible 01:02:55], they are using a lot of the Intel chips. But they’re working on their own Inferentia [SP] process as well. So I foresee that this price will go [inaudible 01:03:07] and move completely into serverless, where hopefully, we’ll never have to worry about, you know, setting up machines or inference units. And so it takes care of itself behind the scenes where you have one rate and you get charged on a per transaction basis. I think that’s the way to go.

    And I have no doubt that AWS will get there. Right now, they use a whole lot of Intel-based inference engines or inference processors. And they have their own new Graviton-based Inferentia [inaudible 01:03:37] Inferentia. Yeah, it’s only a matter of time before most of these will turn into a serverless setup with all of it going to Inferentia.

    I think it was in [inaudible 01:03:53] that they announced the work that they were doing on the Inferentia chip. It’s available today. I just don’t think it’s rolled out with enough capacity.

  175. Stephen

    So their entire machine-learning pipeline is that all gets subsidized over to for the [inaudible 01:04:11] Graviton.

  176. Rahul


  177. Stephen

    Then you hopefully will be able to see things like serverless inference. I would imagine that these are for very low latency applications where you’d want this…I’m still getting over that idea. You wanted your peak infrastructure all the time. But I guess…

  178. Rahul

    I think with any inference model, there are three dimensions of tradeoff. One is, what is the cost of training your model? Okay, but that’s not so much on the inference side. That’s more on the training side. But second is, what is the latency of the results and what is the accuracy of getting these results?

    Okay, so, for example, you might be okay with the machine-learning accuracy being 90%, or confidence levels being 90%. You want the back under 100 milliseconds. But in other cases, you might want that level to be 99%. But that might take you two seconds to get that results.

    So the workload is a function of not just the latency so much, but the latency and accuracy because that’s how much compute you will probably need [crosstalk 01:05:35].

  179. Stephen

    And there’s [crosstalk 01:05:35] returns to accuracy, right. I remember looking at DataRobot, and they give these big leaderboards of models. And there were models that were 2% more accurate that took 10 times more power in terms of inference [inaudible 01:05:52]. I guess it’s a big tradeoff.

    One thing I’m not seeing in the article is, it’s talking about scaling up by the hour. So I guess that is the unit of granularity, right?

  180. Rahul


  181. Stephen

    So you can’t scale up on a per-minute basis. We’re talking hours.

  182. Rahul

    We’re still talking about hours to do this.

  183. Stephen

    Okay. It does feel like what you’d see 10 years ago, you’re paying [inaudible 01:06:15] hour. Yeah, so I guess hopefully, we’ll move towards a serverless framework, or at least more granularity than chunks of hours so it gets cheaper. Because you wanna be able to play with it and experiment with it. But still seems a little bit pricey for that type of thing. You know, if we wanted to build a cat detector for our garden, that might be $176 a day or even [inaudible 01:06:41] of that.

  184. Rahul

    I completely agree. It’s not consumer ready yet. It’s still, you know, like, I guess we find the people are using the mechanism of detecting defect in, let’s say, chips that you’re shipping [crosstalk 01:07:00].

  185. Stephen

    Yeah, where it’s highly profitable and [crosstalk 01:07:02] a problem.

  186. Rahul

    Correct. And you wanted it to run through defect detection on 10,000 [inaudible 01:07:09] and so on. I think it makes sense in those scenarios. But it’s still not quite commodity grade for some of the uses where you’d want to use it.

  187. Stephen

    Yeah. Or if you were adding this to an app you were making for free, then that…

  188. Rahul

    Yeah. Like, if I wanted a content recognition, you know, mechanism that looked at all of my PPTs and extracted a bunch of data from there, allowed me to search through all of my PPTs or PDFs and stuff like that, this would not really be the mechanisms to look at all of those pictures.

  189. Stephen

    Yeah. Well, I guess you could do it all once. And hopefully, it’d make…you have a lot of presentations. So it might take a couple [inaudible 01:07:48].

  190. Rahul

    Yeah. So but it’s heading in the right direction, moving from just seeing what your [inaudible 01:07:59] is gonna be or your maximum capacity, moving to auto-scaling, definitely a step in the right direction. I’d love to see the [inaudible 01:08:08] second ideally, followed by serverless, where you run that…where it’s transaction based.

  191. Stephen

    Then rating it, I think we’ll give it the simplifies. So this is definitely…

  192. Rahul

    [inaudible 01:08:21], yeah.

  193. Stephen

    Yeah. That’s [crosstalk 01:08:24]. And okay. And in terms of the article’s content itself, I think it would have been interesting to know, I don’t know, [inaudible 01:08:36] a little more details into how this would work, how many inferences per second you would want. Just a bit more detail on how this would work.

  194. Rahul

    And I think I’d give this a three. Three seems like a score that fits this particular [crosstalk 01:09:02].

  195. Stephen

    It seems fair.

  196. Rahul

    There’s a lot of interesting stuff that’s come out in the last week. There are a bunch more announcements, but I think this is just about the time that we have. So coming back to the event we have at hand. We have the AWS Partner Summit that’s currently being held here in Anaheim, California. If you’re around, once again, please come and say hi. We’d love to interact with you. We’d love to know what you would like to know so that we can bring that into our future episodes.

    And I will be doing two talks tomorrow. One is on how you can systematically do a 50% cost savings goal with whatever cloud bill that you have. The second one on why you should not really own code anymore, and why code is a liability.

  197. Stephen

    That’s all my IP that I worked so hard to make.

  198. Rahul

    We will have a long discussion about that tomorrow. So yeah, really excited to do that tomorrow. I look forward to seeing some of you here in Anaheim. And yeah. We’ll have a great show tomorrow.

  199. Stephen

    Yeah. Please, if you’re around, introduce yourself and any content that’s publicly [inaudible 01:10:21], we will post that on our social media. All right, this is Stephen and Rahul signing off, and we will see you next week.

  200. Rahul

    Thanks, everyone. It’s been awesome.

  201. Woman

    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 to schedule a free assessment.