Video: From Operational to Cyber Ready: Evolving Your Amazon S3 Protection | Duration: 1652s | Summary: From Operational to Cyber Ready: Evolving Your Amazon S3 Protection | Chapters: Introduction and Roles (28.199999s), S3 Data Protection (118.6s), Modern Data Challenges (273.21s), S3 Data Protection (474.76498s), Modern Recovery Strategies (850.115s), Comprehensive Disaster Recovery (1310.51s), Conclusion and Reflections (1407.935s)
Transcript for "From Operational to Cyber Ready: Evolving Your Amazon S3 Protection":
Hi, everyone. Adam Russo here from Rubrik, our director of cloud specialist. I'm thoroughly excited to be joined here by my friend, John Drake from AWS, worldwide go to market storage specialist leader. John, how are you? Great. Just busy, Adam. You as well. You as well. Why don't you start out first? Tell the audience here just a little bit about what you do at AWS. Yeah, so I've got an amazing role at AWS, super fun. I look after our data protection portfolio kind of across the board. So the idea is we've got a lot of really great native services that we built and developed for data protection, things like SRE versioning, replication. And then we also work with partners like Rubrik that can solve broad customer needs, specific customer needs. And my role is to kind of help us stitch that portfolio together and really help our customers to navigate the right solution sets to protect their data sets, how they can leverage things like S3 for additional copies of data that they're protecting outside of AWS, and really just help them navigate the waters to make sure they're defending themselves against operational issues, disaster type issues. And then, probably the hottest topic we're talking about today, which is kind of ransomware defense and how can we help kind of look at things in a more holistic way. Yeah. So, think on that note too, especially with unstructured data with S3, I think taking even a step back, that's probably one of those areas where I think a lot of users and consumers of the S3 service may say, Hey, I do already have versioning and everything else. Why do I need to have this extra level of protection for ransomware? Is ransomware even getting into these S3 buckets? Isn't that more of a IaaS type of issue that my security team has to worry about? What do you say to those difficult customers and kind of what do you see out there? Yeah, it's tough because when you build one of the best storage platforms on the planet, it becomes mythical. It can do anything. And really what we see is that, we've built S3 around this idea of being striped across availability zones and providing this 11 nines of durability. And operates at such a scale that things that never happen in the real world happen all the time in S3. These are all the types of things that we would need to prepare for, and the system is designed to be resilient across that, which is amazing and a huge benefit for our customers. When we really look at the types of things that can happen to customer data, even something as simple as I delete the wrong object or I delete the wrong file, or I just make a common operational mistake, we need a path to be able to recover from that. All the way up into, how do I start to look at that data state? How do I make sure that I'm doing the right malware and anomaly detection of the data sets? How am I protecting myself against all those different pieces? That still all needs to be in place for S3. So, it's really about looking at the data that's on S3, helping to categorize that data into the right priority for the organization, how easily recreatable is that data set, what happens if it goes away, and then really helping customers to understand that there are lots of things that can threaten data. And S3 protects us against some of them, but it doesn't protect us against all of them. And so that threat really changes over time from what we used to think about, which is, you know, the hard drive goes bad. We've done a pretty good job protecting against that. But now we have a whole industry built around nefarious things to that data set. And so we have new threat vectors that we need protect against. Yeah. I think that's a pretty interesting evolution too. One of the things I love about S3 and AWS in general too. I mean, we're both old storage guys as well too. Well, not that old, but we've done storage for a long time. A But, lot of gray you know, I haven't talked to a customer in a long time in AWS who said, I'm really worried about, you know, a hurricane or a natural disaster taking out my environment. AWS has that completely solved. And that's something I've always noticed too, where because of those huge benefits cloud provides and that resiliency, you just get out of the gate where that was just so much planning and time. And I gotta go build another data center over here to handle this failure use case that some of it ends up getting taken for granted as far as what those modern attack vectors are. And that really is something that we see accelerate a lot here at Rubrik, from a cybersecurity standpoint too, with just threat actors coming in from all shapes and sizes, trying to attack these workloads that are hugely critical for customers now. And one of the things that I think you talked about specifically too, is classifying that data. I think that becomes really, really difficult too. I see with many customers in these large sprawling S3 environments, because it's not a matter of I have one or two buckets, it's half thousands of buckets. It's usually easy to find the two biggest buckets that are the most critical to the business and say, yes, I need to do more there. It's what to do with the other 9,998 of them, and making sure that there's a good benefit analysis. Is this mission critical? Is this tier one? How do you recover? Are we handled via replication? Is that replication just more for an availability use case? I think the one thing that we tend to see, and one that I talk about pretty regularly with customers, is truly moving beyond the definitions of the traditional RPO and RTO of how much data loss are you willing and what's that recovery time objective and really shifting more into what's a cyber RTO. So if you're not worried about just an operational outage, but more of an attack, How are you able to restore quickly? How long is it gonna take? How do you know where that intrusion happened? How are you ensuring that a recovery isn't reinfecting the environment too? That's another big one. And those are typically the things that I see across a lot of customers that are really massive challenges that they tend to try to, they have a lot of struggles with it in answering those questions. Because restoring from yesterday's version is very, very easy. Give or take, depending on the size of the environment there. But knowing that that is clean is actually not going to reinfect that environment. Just adds that extra layer of complexity. That typically isn't thought of when you're building a large data scale application, I would say as well. On that note too, as S3 has really shifted to go way far away from just being a low cost storage alternative, and really now have structured data and critical data on it. What are some of the newer challenges you see customers taking in the world of data protection? Yeah, Adam, you're hitting on one of my favorite things about S3, which is like, well, we've been around forever and it's amazing. And it's really cool to have something that's evolved. But two is just watching the evolution of S3 into more and more criticality from a workload perspective. I think that you've absolutely nailed it back in the day, like how cheap could we make it? How slow could we make it? And really looking at that kind of put data into S3 and sort of archive it off was a lot of the things that we worked with customers on. And more and more and more, we're finding S3 is becoming the data foundation for AI workloads. And so, our push in the product development has been getting up market, building things that are more transactional, things like S3 vectors, which are critical vector database that runs on top of S3. We can embed objects and give that full vector mapping to those data sets, which is really, really cool and really interesting. We've recently released our Iceberg tables, S3 tables, which is a full Iceberg format on S3 natively in that table format. So, we're really starting to see more workloads pushing that direction. We've started to even refactor the way that our service teams work to be able to look at things from not just a storage perspective, but that overall idea of how do we create the data foundation around it. And so as we're seeing those things happen, we're seeing the need to protect S3 evolving as well. When I look at S3 as a whole, it's about 5% of the S3 footprint across AWS is what we would call properly protected. And 90% of the data has no protection on it at all. And so, when we talk to customers, one of the very first things we do is we have that conversation that we had. We also look at the workload and say, Look, your workloads on S3 have changed radically. This isn't just a backup copy. We're now using it for these AI workloads, these critical data ETL workloads, and your protection strategy is still the same as it used to be. And so how can we now enhance the protection strategy to be commensurate with the type of data that you're storing on S3 so that we have the right protections and that you really can adhere to your internal policies that you may have around data protection as well? Because I think sometimes SRE gets left out of that conversation when we're thinking about being compliant and complying with internal policies around backup and data protection. So I think you hit on a good point there too, with these conversations. So I think that relates back to something that everybody's holds dear to themselves too, right? The shared responsibility model. So AWS is responsible for the security uptime of the data centers and the data in it. And then it's each customer's responsibility to protect their own data on top of it. But one thing I've noticed kind of similar to this discussion as all these S3 buckets are going and owned by all these different application teams as well. So it almost ends up being a meta shared responsibility model within the organizations themselves in the sense that this app team, are they responsible for it? Is the platform engineering team responsible for it? Is security responsible for it? Because we're talking about cyber. And really who owns what, where that is part of that. It's kind of everybody's responsibility and it has to be everybody's ownership to kind of get to that next level of what happens if this bucket doesn't go away? What happens if going back to even those AI workflows as well with Agenic AI, not only do you have to worry about just somebody running a script against the wrong bucket and wiping out half of it. Now you might have Agenic AI doing that for you, which is another one of those failure scenarios that ten years ago, nobody was thinking of. And that really goes back to, I think that criticality of these large scale data sets and ensuring they had the right level of protection based on their value to the overall business and the cost of that downtime related to it as well. Yeah, Adam, you bring up a really good point. And I was on a call with a customer just before this. And there, we were talking about doing some cost optimization, which is a big topic we always get into. And it was really around this idea of kind of going up the chain to understand who owned the backup policies, who owned the data retention policies, who owned all these different pieces. And so you bring up a really good point, and it was the spider web of involved personnel to make the decision of how do things need to be protected? What can we do to shorten retention periods or change retention periods or ensure that we're in compliance with the original intent of this policy? It became very, very complicated very quickly with this healthcare company who's tightly regulated and has tight standards on that. Esther is in the center of that conversation, but it became much more of a business process conversation. So, bring up a very good point, which a lot of our consultative discussions with our customers are not about, Do you have versioning on? Or, Do you understand how a backup works? It's much more about, How do we help you navigate all of these higher level procedural policy based discussions that you need to have internally to even understand what to set the numbers for in those policies and how much of the data this was also about vaulting how much of the data needs to be scanned, immutable, air gap, copy that's isolated in a separate account? They were putting a tremendous amount of data in that space, and they were asking me if I felt like it was right. And the answer to that was like, is your policy? How do we help you uncover that? And so that's really been a big part of the conversations we have with our customers are understanding not only how the tools work, but also how does your organization work to be able to meet those requirements. I think related to that too, I'd to something we talk about regularly too. 32,110. What does that mean? 32,110. On the on the spot, I'll I'll rip through it. So, you know, the the idea of this is kind of expanding on the older concepts of backup. Right? So, you know, the idea would be we wanna have, you know, three copies of data in addition to your primary copy. I want two of those to be in separate spots. And, it's a little bit separate from the old concept of two different types of media. We're talking of serially about where things are. But how do we get it in two separate locations? At least one copy of that, we want to be in a cyber vault, which is off in a separate account. It's immutable. It's air gapped. It's got those protections against it. And then we also want to have that additional copy that's ready for rapid recovery from a disaster recovery standpoint. Zero is actually likely the most important part of this whole equation, which means we're regularly testing to make sure that we can actually recover. And I think that in that space of can we recover, that is an area where I've seen customers really changing their approach over the last few years, which, how can I recover? It used to be the test was, I will randomly pick a file and then we'll see if it'll restore. And if it does, then my backup's good and we will call it a win. Much more into this idea of, well, recovery really means I need to have these set of applications up and running. I need to be able to serve out an actual outcome to my users, which means I need to have all of the dependencies satisfied for an entire group of applications to be able to serve up that outcome. And, maybe on top of that, I need to actually understand what my minimum viable company is or what my minimum standard is to be able to run. And I need all of those applications to come together to say this is the critical thing. I was meeting with a financial services company a couple of weeks ago, And their standard was, how do I process a credit card transaction? That's my minimum viable standard. Can I get a credit card swiped and it hits the ledger? And looking at the number of systems across different geographies that all needed to be in place to do that one simple thing that they felt defined their minimum viable bank in this case was immense. And so for them, a recovery test is going from an event, say, called a ransomware attack, to being able to process a credit card transaction within that recovery window that they defined. That's what their recovery test is. A far cry from, Can I get an Excel file back? Which used to be the standard. So that's been kind of the evolution that we're seeing in terms of what does a recovery test mean. It's really been pretty transformational in terms of those standard and how that zero equates. So, long answer to kind of what our 32,110 is, it's a great framework to think about all those different pieces that come together, but it's also a modular framework. And I think this is the part where customers get most caught up and where we see things like, what can we do for an immediate cost optimization? Not all data needs to be treated equal. So, Adam, back to your earlier point about data classification, the full three two one one zero may apply to five or 10% of a total data set. Your mileage will vary. Everybody's environment's different. But it's a very small subset relative to the total data set, which may only require a single one or may just require in place on storage versioning because it's an easily recreatable data set. So working together to understand what that requirement is, assessing the environment, and making intelligent decisions to meet your internal policies for protection can help you take 32,110 and make it a very practical framework to apply to your overall dataset. Yeah, that's huge. I think the practicality and kind of not going over the skis of like, I gotta do this for everything and go from one copy to a 100 copies of everything in environment. That's too much of an apple to bite off. It is not necessary either. Right? But I still really love the Xero because I think that is the most critical one of all of us too. And I think fundamentally, that's one of the ones that a lot of organizations don't particularly think about, either until they have a bad day and then they go, Oh, yeah, this wasn't a good experience. Let's go improve this afterwards. Or I think the other one, especially related back to S3, these applications are becoming so crazy large and fast in the velocity and the number of objects that getting to that minimum viable, there's some physics involved with it as well too. There was a health tracking customer that we recently worked with as well, where they had a similar They have all their user endpoint data and trying to talk through what the minimal viable product would be if they were hit with a ransomware for their large S3 bucket, has a lot of parquet files and etcetera. All this data is constantly updated. They have data for multiple years that they're running analytics on. Their end users can see all of that. But in the event of they're all down, they don't need the last twelve months worth of usage, right? Their end users are opening up an app and they wanna see today, last seven days, last fourteen days. So that needs to be available right away. And this is a multi billion objects. So you have to pick out just those most recent objects, restore those first and then everything else afterwards. That's a challenge, especially compared to the old days of that Excel file or my database is on this server. As long as the database comes back, we're good. And just that physics challenge with just the scalability we're seeing nowadays and just how fast organizations are developing and growing and the amount of just data growth we're seeing too. So, I really love that example. Yeah, absolutely. Absolutely. I think the other thing that I'm starting to see is this kind of and these are areas where I think we're really cutting edge and helping customers really think about this problem, which is this idea that in a lot of cases, we historically thought about recovery in place. How do I get this stuff back? And in this kind of new climate, we really have to think about standing up in a net new clean room or a net new environment that is known good, which generally means we've built it from scratch. And now we have to lay down all of the foundational work to be able to make it run that application. And so, how we can work that into the overall process is another huge part of this. Orchestrating the setup of the environment in relatively straightforward for customers trying to recover an AWS environment in AWS. Exponentially more complicated if it's something that's on prem that needs to be recovered into cloud. The benefits of AWS for CleanRoom are great because it's on demand. We can scale. It can be exactly the right size it needs to be. Everything's on demand, but it does have some challenges in terms of orchestration and laying that out. So, I think the more we can work together to help customers define what that looks like and build that out, they can take advantage of those benefits and then get that process of practicing what a recovery motion looks like, so that it can happen in the timeframes that expectations hit, if and when there is an issue. Yeah, I couldn't agree more too. And that's one of the exciting things I think we start to see is, as we keep having these conversations customers too, it's okay. Here's how we can restore the data, maybe some of the infrastructure in the event of a cyber attack. We can get the clean point of recovery quick. And okay, you're gonna deploy all of your infrastructure via infrastructure as code. And then that's gonna lay everything out. And then you get to that next level of, well, what if the attacker goes in your DevOps environment, GitHub, Azure DevOps, wherever you might have that. Right? If that's unavailable, well, then how do you run your deployment pipelines? So that's even something where we've taken on as well too, or, Hey, we can back up and restore GitHub. So then we can restore your Terraform pipelines, CloudFormation, wherever those live, to lay that framework down first. So then we can restore the data on top of that. And then even branching off beyond that even into identity too, which is another big one for us. Because often identity is the path of entry for a lot of these organizations. And if your identity is unavailable, it's hard to log into anything to do much of anything at that point. So we've been being able to take that backup and restore there. So a lot of these conversations with that minimum viable company that we have, it's not even just a matter of that data anymore. It's the layers of that infrastructure and how you would get that back up and running. And fortunately, that's something we're able to partner on too, where we can kind of answer all levels of the stack, including even things that are still in the data center as well too. So really exciting to be able to really answer those needs for our joint customers. Yeah, absolutely. And one thing to kind of touch on another point that you made that comes to mind is really this idea of, as our customers are pushing into AI and AgenTeq and pushing into S3 Vectors and pushing into these new offerings, just understanding what data would be recreated versus what data needs to be protected and kind of helping navigate all that. I was in some very lengthy conversations with the customer about S3 vectors. Was it faster to redo the embeddings? And was it less expensive to do all that processing and then get a more revised embeddings map as a benefit? Because the restoration of things is moving around and is it going be better to embed again? Or is it better to protect that data and have a backup and restore as was? So it's very interesting to see as things are evolving, really helping customers understand what needs to be recreated, what needs to be backed up, how do those things come together, how do we think about them differently and separately. It just seems to be much more complex than it used to be in terms of really navigating that. I think that's an area where I'm always excited when I get to do those exercises with the Rubrik team because the expertise and the knowledge behind how to help customers navigate those waters is it runs deep within your team, Adam. And so it's always fun to get to do those projects together because I learned something. The customer learns a ton and we really get to help come up with a great outcome for that customer. Yeah, I couldn't agree more. And I think that's one of the things that is really great because customers are constantly challenging AWS to constantly challenging Rubrik because they're constantly innovating, chasing after what's next. And it's always great to see. And it's great to even take that kind of step back and answer the challenges kind of early in that cycle too, before they get to that point later on in their development cycle or where it's production and being able to get some of these via classification, how reproducible is the data, having them think through some of these things earlier in that development process so that it doesn't become that emergency later on. But it is great too, just seeing all the new innovation all these customers are going at and the new challenges they're seeing day by day. Definitely excited overall. Yeah. Our builders are so fun to work with because it's like, know, they're, everybody's so smart. All of our customers are so smart. They know what they're doing. Have you thought about it this way? No, we haven't. That's great. So it's super fun and engaging to get to come together with our customers to solve these problems. Yeah, I couldn't agree more. Couldn't agree more. Well, John, think we're out of time here too. So thank you so, so much for the time today. Great catching up. And everybody who joined us today too. Hope everybody has a great afternoon. Absolutely. Thank you, Adam. Thanks.