The Hacker Mind Podcast: EP 69 Self-Healing Operating Systems

It’s time to evolve beyond the UNIX operating system. OSes today are basically ineffective database managers, so why not build an OS that’s a database manager?

Michael Coden, Associate Director, Cybersecurity, MIT Sloan, along with Michael Stonebreaker will present this novel concept at RSAC 2023. You can learn more at dbos-project.github.io.

So while I was editing this podcast on self-healing operating systems, I was reminded of an article that I never finished for Fobes.com. It was on a project that Dan Kaminsky presented at Discourse 2019. He died prematurely on April 23 in 2021. Maybe I’m thinking about Dan because of that anniversary. 

I first met Dan when he was literally saving the world; okay, at least saving the internet as we know it today by disclosing to the major ISPs in the world a flaw he’d found in the Domain Name System or DNS. That’s the translation of Google.com to a series of numbers that is the physical location of the server responding to internet requests. Dan found a flaw that could have crippled the internet. For some reason, shortly after, Dan gave one of his first interviews to me in the studio at CNET.

Over the years, Dan continued to do really great things. Like he created an app for people with color blindness called the DanKam. Using a variety of mechanisms, it allows the color blind to determine the colors of objects around them. 

One of the last things Dan presented was a time machine. He called it Time Stone. It was not a literal time machine, but a way of capturing the software development process by recording intervals and storing them in the close. Then, when a vulnerability was discovered later on, a developer could go back in time and find the moment the fault was introduced into the code. 

In the Forbes article I never published, Dan said quote- “The reason that software is so inexpensive, we actually are very good at estimating development time. But if there's a problem, the problem time is exponential with amount of crap we're working with and we are working with a lot of crap. I am tired of being Sherlock Holmes." - unquote

The idea that you can isolate a change and then fix a specific vulnerability harmonizes with the topic of this episode: self-healing operating systems.  Rather than restoring from a backup the entire operating system, which gets you everything up until the moment of that last backup, then you still have to add  in what’s missing. What if there’s a way to isolate and specifically rollback the one or two thing that changed? Kind of like a time machine, only it’s much more. It’s a radical rethinking of how we even view our current choices of UNIX-derived operating systems. And in a moment I’ll introduce you to someone who may have found the next generation beyond the current UNIX systems we use today. I hope you stick around.

[Music]

Welcome to The Hacker Mind, an original podcast from ForAllSecure. It’s about challenging our expectations about people who hack for a living.  I’m Robert Vamosi and in this episode I’m exploring a novel database operating system that promises to be disruptive to how we mitigate malware today. 

[Music]

VAMOSI:  What is an operating system?  It manages all the hardware and all the software on a system. So what do we mean by managing all the hardware? It manages all the resources that are needed, such as Random Access Memory or RAM, and hard or solid state drives. And software? That’s your applications, your web browser, you streaming media. This might seem obvious, and yet … there’s a point.

The first digital systems simply ran software, meaning it ran one program at a time. That program, then, handled all the issues of hardware and software. No need for an operating system. Over time, that provided not scale. For example, you’d need several different systems, each running just one program, to accomplish a task. 

In the 1950s we started to get early operating systems, and these included supervisory programs that helped manage the data coming in and going back out.  This started to get us to having more than one program run at a time. 

By the 1960s, it was clear that multiple programs required time sharing of the central processor, the CPU. One of the very first operating systems was known as Multics, because it allowed multiple programs to rum on a mainframe system. This was superceded by UNIX, which survives to this day in one form or another.

The point of this? What if we could further evolve the basic operating system beyond just managing resources but also now mitigating any malware trying to make changes. In other words, what if we could have an operating system self-health itself against mischief? 

So, with RSAC 2023 around the corner, I met with someone who’s speaking on Thursday morning, and also launching a self-healing operation system for public discussion. 

CODEN: I'm Michael CODEN. I'm currently associate director and co founder of MIT cybersecurity Research Consortium, which is called cybersecurity. At MIT Sloan, which is cams cms.mit.edu. I'm also a senior advisor to the Boston Consulting Group BCG. 

VAMOSI: Michael has quite the pedigree. 

CODEN:  From 2016 to 2021. I was the head of the cybersecurity practice. At BCG. I did a lot to build that practice over six years. Prior to that, I was with an Israeli software company instead of a security software company and practitioner dealing in software for critical infrastructure, working with companies such as GE Healthcare, Motorola's cellular communications network, industrial automation companies, like ABB Honeywell Rockwell Schneider Electric imagawa oil companies like Shell Oil, and so on. And, and that did that for 13 years and because of my involvement in critical infrastructure, I was asked by the White House to assist in developing the NIST cybersecurity framework in 2013. Then the Obama administration and in 2014, we published that that's when I went back to MIT to a former classmate, actually the guy who lived across the hall from me in my dorm, and we co-founded the cams cybersecurity at MIT Sloan. We have about 23 sponsors for that. companies like Verizon, Google, Microsoft, State Street Bank, mutual, BNP Paribas, some oil companies, and and then through our work at MIT Sloan, we also get very much involved with the Computer Science and Artificial Intelligence Laboratory which is CSAIL. And that's where I met Mike Stonebreaker. And that's where the story of dB the database operating system, this revolutionary upside down operating system technology begins.

VAMOSI: So on a laptop or even a phone, there’s an operating system that allows you to laid applications and run them. On a chip, there’s what’s called real-time operating systems or RTOS that executes commands. And in the cloud, there are microservices that perform workloads.  So, before we get too far into this discussion, let’s define an operating system. What makes something an operating system?

CODEN: Great, great way to start. Great question. So an operating system provides a few basic services to an application file structure, Inter Process communications between different processes, indications in and out of the file system like TCP in and out of the machine like TCP IP, something like that. And scheduling and the original operating systems that were developed like in like 1869 7071, Unix and and shortly thereafter Linux, they ran on a single CPU, and they would have a few kilobytes of memory and what that does, they can so I actually I'm old enough that I visited Bell Labs and I've seen involved with Ken Thompson and Dennis Ritchie developed Unix on a PDP 1145, which had a single CPU, and somewhere between 64 and 128 kilobytes of memory. So that's what an operating system was supposed to do. And now, what has happened is we've got much more massive hardware, much more massive memory, much more massive applications. And so the operating system is being able to ask when the system is being asked to manage resources that it was never designed to do. 

VAMOSI: What Michael is suggesting is that we’ve outgrown our current operating systems. We’ve reached some logical limitations on what we can expect going forward. So a new way of thinking about operating systems is necessary. Hence, the database operating system.

CODEN: And, in fact, the way that the database operating system concept was born, was my colleague, Mike Stonebreaker, is a professor at MIT. He won the Turing Award for developing a lot of relational database technology and invented Ingress, Postgres Illustra, which can use Vertica OLTP and other relational database technologies. And he was talking to another one of my colleagues and in this database Operating System project Matej is a Hurayrah wrote is the creator of Spark and, and he is a co-founder of Databricks. And he was discussing presenting a paper, I think, and Mike was in the audit, although they know each other very well. About how there were millions of different state variables that needed to be kept track of by the the operating system. It was getting so cumbersome, that they actually exported all the state variables into a Postgres database in order to manage the operating system. And so this kind of created the spark that Mike starts making. And this is three years ago and now we're gonna we're I've been working with him for almost two years and in the team of 20 has been together for close to three years. And we've developed even a prototype approval in these things that I'm going to talk about. 

VAMOSI: Michael and Mike Stonebreaker will both be presenting a talk on self healing operating systems at RSAC 2023 in San Francisco, CA. And whether or not you can attend his talk, he’s very interested in getting your feedback. You can go to dbos-project.github.io for more information on this open source project, and to test out the concept itself. 

CODEN: Basically, the fundamental issue is look at that early Unix and Linux operating systems that were providing these very basic services that are still the the only things we really need from the operating system today. The operating system has grown because we want to use multicores. We add Kubernetes via containerization. All kinds of security protections, different things. It's become a massive piece of code, and a lot of bolt ons, and it's actually managing a million times more state variables than it had to manage four years ago. So the as Mike likes to say without me saying it's a really word. That's a database problem. So the operating system has become something that provides these very straightforward operating system services, but it's also become a very inefficient database management system. And so that's where we came up with the concept of why don't we do away with the operating system and put a database management system directly on bare metal or the hypervisor and then run the applications in the in the database as stored procedures, and all the log files are stored in the operating system. So application logs, database management logs, and OS logs are all structured in the same format stored in the same database, which is the operating system.

[MUSIC]

VAMOSI: Just want to let that sit a moment. A database operating system. A database to manage all the services and resources. There are some immediate benefits to this. One of which is performance. 

The database guys point of view, this was like, the performance is amazing, right? When I heard about it, I said, Oh my god. With SQL queries, I can do anomaly detection and detect cyber attacks. With simple, really fast SQL queries. And we actually did some tests, comparing data we've gotten from three partners, commercial companies that have provided us with sample data and applications, and we've run tests of our data versus the most used external analytics engine, which will remain nameless, but it's awfully expensive. 

VAMOSI: There is something else here. For security products today, all those endpoint detection systems take some time to filter through all the subsystems. With a database operating system, you’ll known in milliseconds if you’ve been attacked with malware. 

CODEN:  And whereas extracting all the log files, converting them to the same format, putting them into the analytics engine, running the analytics engine, you're about a four hour delay. Until you can detect an attack inside the database operating system, the same logs, we ran our SQL queries, and we're talking like, hundreds of milliseconds to detect the attack, basically converted these very complicated analytics engine rules to simple SQL queries. Much easier to write, and much faster to read. So that was the first thing that caught my attention. And the second thing is, databases. We have the ability to roll back to a previous state. 

VAMOSI: And there’s an additional benefit. As a database, you can roll back to a previous state. Not restore from a backup, because other parts of your system may be doing other things. Rather, you can rollback a specific part of your database operating system, leaving the rest in tact. This is a novel approach to malware.

CODEN: Right, so So rather than restoring from backup files, which are stored somewhere, we actually we have a function within a database operating system concept. It's called prominence and using a simple feature of typical database called Change capture, you keep a log file of every data elements is touched and what happens to it and and then you can rewind the through this lock file or this provenance file, and you can, within a few seconds, get back to the previous tech state. So all the logs are kept in the database, SQL queries the tech the attack, we let you know about the attack, block the attack, and then we're rewriting the operating system to the pre attack state. And you're offered running business continuity in seconds. And, and then to carry this to you know, just to add a sidebar on this. My my personal work, personal work. My other work that I do is some other hobbies that I have are worrying about the balance in cybersecurity of resilience versus protection. And the fact that what I'll, what you're, you're building your, you just can't protect yourself against everything. And so, companies have to think more about how can I quickly recover? And that's what really attracted me to this concept of the database operating system. And it also goes a little bit further and I've been talking actually to members of Congress about this. We can come up with a cyber deterrence concept, which would be determined on the basis of denying benefit. So, if if my system is attacked by ransomware, and I can detect the attack and recover in seconds, and the adversary gets no benefit from having a technique. They'll eventually stop.

VAMOSI: The anti-malware aspect of this is pretty interesting. 

CODEN: We're going to have all the rest logs. All the database management system logs, all the various application system logs all in, in tables in the operating system. Over time, that's going to get to be a pretty big table. So what we do in the open source systems is we spool that table to a column store. The one we use is Vertica because Mike Stonebreaker developed it, but you can use open source ones like redshift or snowflake. Doesn't matter. But the interesting thing about that is that these column oriented data warehouse DBMS are extremely fast at searching. And that's what gives us the hundreds of milliseconds to detect an attack time and it's being done not literally not in the operating system but adjacent to the operating system in a column store. database. So it's because the data you want to look at over long periods of time.

VAMOSI: Anyone who has worked with databases can tell you they can sometimes be maddening. Now we’re talking bout an operating system that keeps track of everything in a database?

CODEN:   we did some tests against one of the most, as I said, against one of most well, we got to test the license against one of the  , most expensive analytics engines. And, you know, by the, the conversion of the logs was so complicated. The writing of the rules is so complicated, but it took between the conversion time and the processing time hours to detect an attack. And because all of the logs are in the same log file, so think about it. Today when you send your your logs to assume you've got many applications running, and they each have their own Wattbike in the database, has its own log file, or multiple databases. And the US is still like, and they all have their own timestamps. Right, okay. So, in order to determine a tag, you've got to merge these and sync it  Yeah, because I want to know that Rob was trying to get to this data from New York and from Bosnia at the same time, right. And, and so the fact that these are all restructured in the same file, you know, in Tirana, logical sequencing on the same table, sorry, incredible. That's what helps make the searching and the detection so quick.

VAMOSI: The granularity of this with regards to malware is pretty cool. It’s like the original operating systems were built to simply run multiple programs. It wasn’t until we connected these machines up to the internet that we started to see abuses on a large scale. Now, with a self-healng operating system, we’ve matured to the point where we can deal with these modern problems

CODEN: I mean, security is built in. It's not patched on. Pretty much new code. The unified logging is really, really important. Files, messages, database scheduling, it's all in one place. The built in rabid attack section we've also developed machine learning tools that look at the logs and generate the SQL code. So we can actually, the system can learn. We've been experimenting now with with ml. So system matching learns what to what's an anomaly in addition to all the hundreds of rules that are typically written into a sim and the built in data governance, or GDPR, the CCPA California Consumer Protection Act, and then the fact that the programming is is easier. Oh, one of the things so we developed a time travel debugger. 

VAMOSI: And now you see why I was thinking about Dan Kaminisky’s unfinished project. This is essentially the same, but in a database operating system instead.

CODEN: Okay, so what's the worst bug that you can have in your software? The one you cannot reproduce? We call those Heisenbugs. They're usually in bad race conditions. Yes. So your, your customer calls you up and says, here's this code. It has a bug and you can't reproduce it. So what we can do because of Providence, is we don't try to reproduce the bug. We roll back the system and re execute and we're able to debug software and a fraction of the time that that it takes using conventional approaches.

VAMOSI:So we talk about it being more secure and you make some reference to ransomware. In particular, walk me through how an attack would be noticed or stopped using your operating system.

CODEN: So if someone were able to get into the system, as I assume eventually they will be able to the sequel rules that are scanning the logs should be able to determine that there's a ransomware case there's probably a lot of data changes and a lot more data changes going on than the normal. Somebody's trying to say encrypt the whole database or exfiltrate the whole database. So that should be pretty easy to figure out also, we'd be able to tell that the actions are coming straight from an external IP address that we're not familiar with. The same rules that you would have in any sim system would be written in SQL in the operating system. When we detect it, we would alert them to shut down things. Stop, stop what's going on. It would be then up to the IT organization to block that attack. We're not going to block it ourselves. We alert and then somebody has to determine what's the appropriate thing to do to shut down the airplane in midair. The then at that point, there is a simple script which says once the attack has been blocked, which is rolled back and you specify the number of minutes you want to rollback and the the provenance will then roll back the entire state of the operating system, the entire state of whatever database has been manipulated back to the original where they were to 18 minutes ago.

VAMOSI: So, again, this is different from a backup because a backup is going to restore the good and the bad that went on. Whereas you're rolling back just the things that change because you've got that log file telling you exactly what changed and how it changed.

CODEN: Exactly. When was the backup made? Was it made today, yesterday or last week? So the ability to just roll back and then if you want to you can even look at the tape. Take what happened between say 18 minutes ago and now and you should be able to extract the valid things that the system tried to do from the invalid things and even get yourself closer to real time. Yeah. But you're you're you you said it very well. You just wind it back here at the most recent possible accurate state.

[MUSIC]

VAMOSI: So we talked about Unix as a definition of an early operating system. There's the commercial side of it, you know, Mac OS, Windows, Android. And then there's the other side, the embedded side where you've got real time operating systems. And then from the commercial, I would abstract that we get microservices and containers and the whole cloud structure. So where does a database operating system fit all or specifically somewhere?

CODEN: So one of the things digging down a little bit deeper is that many memory distributed databases today are multi node multi core and high availability in their basic nature, so containerization is not required. So you develop your software, you compile, we build a programming environment, I call it a compiler, but I'm told now to be that's a net sh term. It's a programming environment. It's called EPR. It's open source. It's available on our GitHub site and takes your Java code and breaks it into a series of functions. That becomes stored procedures in the database to bring the program to the data, which is the most efficient performant way of doing it. And oops, I forgot where I was going Interjection right. So, so after developing your program in the programming environment, it's immediately deployable. You don't have to go through that extra step of containers, manual containerization. So you actually can deploy software significantly faster. And the whole surface area of the operating system is much, much smaller. So by by just by virtue of that, it's going to have your own vulnerabilities.  So in the cloud environment. We've actually run on the MIT supercomputer which is 9000 cores. And the head of the MIT Supercomputer Center at Lincoln Labs is also part of our team. Jeremy Kepner. That's a fabulous guy. brilliant guy. And in the so for for cloud environments the most. The way our programming environment compiles code is serverless. So it's similar to if you were using say AWS lambda, but it's far more efficiently structured and gives you the C unified logging. And it's much faster in comparison tests against lambda; the performance of the stored procedures in the database has been about 10x faster.

VAMOSI: So the real time operating systems can benefit from this database structure as well and how might that be?

CODEN: They definitely could benefit for several reasons. Security. Clearly one of them and the other thing is that real time operating systems are to a large extent transaction. And this is just transactional, but was in its very nature. So it'd be more efficient and faster and can significantly speed mealtime applications. The reason I don't jump to that initially, is having this industrial background, but I know that there's like a 20 year life cycle on real time operating system type equipment and the equipment. It's a much longer lifecycle to get into. But it's definitely a place where we see significant benefits of work. In areas like manufacturing, in areas like medical technology, areas like defense technology, because of the increased security and the ability to deal with the transactional nature of those systems.

VAMOSI: I would imagine the footprint is significantly smaller to significantly. So in an embedded system, that would be ideal because it's resource challenged, memory challenged, etcetera, etcetera.

CODEN: We need to develop a relational database that's small enough. So that would require some development but memory is cheap these days. You can if you can get this increased security for a little like $10 more memory. It's probably worth it.

[MUSIC]  

VAMOSI: So given the efficiency of this new operating system concept, given it’s ability to stop malware the moment’s detected, what industries does Michael think might be the first to adopt this new technology?

CODEN: Or in the earliest eras we believe we're going to be in financial technology and ecommerce. Those are the applications that a lot of Greenfield applications are being generated all the time. A lot of personally identifiable information is being collected. And this both protects that information and there's another really important benefit of provenance. You know, we're all familiar with GDPR and the concept of the right to be forgotten. Today, as far as we can tell, there is no good way of finding out where all the data by PII has been sent in existing software systems. But our provenance capability tells us exactly where every piece of data has ever gone. So we can go back through a database operating system, system, and, and collect audits when you say, you say I want to be forgotten. We can go back and we can find your data at every place we've ever sent it and delete it and prove that we did that. Which is also in the province flag. So the governance the data governance capabilities of this system far exceed anything that's available today as far as we can tell.

VAMOSI: Michael mentioned financial services. And the idea of self healing is very important, because, as you mentioned, they've got a lot of PII that they're protecting, I would imagine, medical and other industries would be attractive as well.

CODEN: Very much. And that also gets to the real time operating systems, the IoT devices, embedded medical devices, but most of our embedded medical devices today are transmitting their data to cloud environments. So we'll probably start in the cloud environment and then work our way down. And, but there's also a lot of work that I've been peripherally involved in, in, in patient systems. So a lot of hospitals are. So I mean, if you just deal with hospitals, well, first of all, the ransomware problem is huge. Inside hospitals, so you know, if we can get this database operating system approach used in the systems that are saving electronic health record systems, I think that would be a massive benefit. Because that's what's been shutting down hospitals. The next generation of catastrophic hospital events is going to be when the medical technology systems, blood analyzers and the pharmacy system and other monitoring systems are attacked. A little bit more difficult because those are for a new software to get into because those have to be typically FCC or yeah, that FCC, FDA. FDA. Approved. But yeah, I think that's that's, that's really important. But there's also a lot of development now and trying the hospitals have a huge financial cost problem. And there's been a lot of movement in terms of developing systems where patients can take care of themselves to some extent they can go online, they can make appointments, they can do tele visits, they can get their bill online, they can pay their bill online, they get their medications online, they can get the I just have my hip replacement. I get my exercises online. They send me every, every week as exercises I'm supposed to do and it's so I think there's a lot more of that happening. That's new applications that are Greenfield applications and they would be a good place for medical systems providers to try a new operating system.

VAMOSI: So three years ago, Michael’s colleague had an epiphany and that made sense, as all novel ideas do in retrospect, they're incredibly obvious and simple. I'm just curious, has this concept come up before in the history of computing and what were challenges with those other earlier ideas?

CODEN: As far as we know, it's not that we've been publishing papers for two years now. And then we don't see any competing research. And they're all available on the GitHub site de vos dash project GitHub, that IO. And what what's happened over the last two years is two and a half years now is we've built prototypes in the software they were open is all open source and the first prototype versions of the first product prototype did on Postgres, which are very well known database, relational database management system, and it is also open source, Postgres is open source. And so we built several different file systems, inter process communication and scheduling systems on top of Postgres. You can run applications and the APR in the programming environment will allow you to compile conditions that will run on top of Postgres, you can actually try it all the functionality using this open source software. The wrinkle is that Postgres is kind of slow. So the performance of that operating system doesn't compare with conventional Linux Kubernetes certainly yet the trade off there was ease of use and good security versus performance. So what we needed to do was find a faster database. And what one of the database systems that Mike Stonebreaker had developed. It's called bolt dB, and it's the main memory OLTP multi core multi node high availability database, that's blindingly fast. The problem is he started a company and the company he's no longer involved with, but the company decided not to open source the database. But we did get a license. And we built all the same stuff that we built on Postgres, we built a multi beam and ran it on VoltDB. It's every bit as fast as Linux. Some things are a little slower, but some things are a lot faster. So it has all the performance of any conventional operating system. And what we're looking at now is because the OLED TV is not open source, we're looking at a foundation dB, which is open source and is also very fast. And it's also a main memory database. And we're looking to convert everything and convert Bill foundation DB into the database operating system approach. So there will be a fully open source version that will be available to the public. So I guess that that's the performance that we wanted to have.

[MUSIC] 

VAMOSI: I personally have a theory about secondary and tertiary innovation, that you need that first level to saturate fully before that second level comes. And then the third like you don't realize there's a problem until you have all of the saturation out there and you recognize Oh, there's a problem and then you build that secondary, and then maybe even tertiary innovation on top of all of that. I’m wondering if we needed the early operating systems to reach a maturity point where we saw problems with them, and then this idea came forward. It's like, well, database, we're really dealing with the database problem here. It's not an operating system problem. And thus the genesis of this perhaps.

CODEN: Exactly. Okay. Exactly. Right. Yeah. Seems like there's probably some good analogies in other areas where something you know, Unix and Linux were great for many years. But as the the available hardware became more extensive and less costly, and as the applications became more complex, they've now outlived their usefulness in our opinion.

VAMOSI: So, if I’m understanding this right, the database operating system could be expressed as a system of tables.

CODEN: Exactly. Right. One of the revolutionary things about Unix Analytics was the use of files or inter process communication and scheduling and then different services OS services. So, the change from the previous generation of operating system versus is a really novel thing that you can send files in between different processes. And what we say is that that's kind of old news. That's an old fashioned, slow and inefficient way of doing things. Our new mantra is that everything is a table. So let's say I want to build a file system. I have a table, a block table. I have a table for directory, I have a table for permissions. It's just a few tables. And then using SQL instructions, I query these tables, find out what's available and insert my files into the file system and I put the information into the tables. I want to do inter process communication that just put an entry in a table communications table, and the other process takes the entry out of the table. Scheduling is a table full of resources. And if I want a resource, I look and see if it's available. And I put my initials next to it, use it and when I'm done, I take my initials away and somebody else can use the resource. And so this, this whole concept of, you know, changing files to tables, is incredibly efficient. And easy and fast. And so all the operating system state being in database management system tables, is one of the things that allows us to recover the whole operating system to a previous state. Absolutely know at any given moment what the entries in all those tables are.

VAMOSI: I wonder if an appropriate analogy might be XML. Interesting. You just abstract the data and share the most important parts and then how its formatted at the other end is up to the receiving application.

CODEN: I started off talking about how it has always grown, you know, right order by six orders of magnitude over the last 40 years, that it's a database management problem. And then have a diagram that shows that the you have instead of user programs sitting on top of an operating system and database sitting on top of an operating system and the operating system, sitting on top will actually they're all sitting on top of Kubernetes and Kubernetes operating system and there are many operating systems and they're sitting on many kernels. And that's on the silicon and we show that the diagram that shows that we have a minimal kernel, we have it distributed in a multi core multi node OLTP high availability database, and then we have the sequel services that are the operating system services and then they use their applications around on top of that. You can have a second database next to the operating system database. So a lot in our study of the way people are developing their systems. There's a lot of data that they can't take out of existing databases, or that they're a database schema that they don't want you so we call it the application data. And so what you can do is you can have a database operating system in which all the operating system states in all the log files are in the database operating system. But the application data like the banking information or the shopping information is in stolen Oracle or some other database, whatever database they're using. And so that's sitting side by side on the same kernel. And, and we think that's that's one way that that people would would be able to use the system,  And you know, the fact that we've, we're able to build the prototype projects available to people with tests, we'd love to get people to do that. We'd love to have them come to our session at RSA.

VAMOSI: Like any novel idea, Micheal is aware this is an uphill battle. To challenge the incumbents, not only the traditional operating systems but also the whole security industry as well. 

CODEN: One of the reasons I left -- retired -- from BCG and enjoying this group of vagabonds is I think this and replays in this have the same impact on operating systems that Unix and Linux had on Windows, right? It's become a major force and I think this is the next generation of general purpose operating systems. But we can address the I mean, that's a long, that's a decade's long process. It's not something that's going to happen because nobody's going to take all their existing software and convert it. We don't, we wouldn't expect that. And what we're looking for in the open source, open sourcing the software, throughout my team and Stanford, because we're looking for people, to be users and developers out there to come back with us and say, and help us either just tell us what features they'd like to see in it that we can develop or help us develop them in any other open source project. Or a project. I don't want to call it. It's definitely not a product, it's a project. I use the word revolutionary and people kind of look at me like you're you know, you're hyping it but yeah, I just wrote an email to a very good friend. I said this is the most revolutionary thing I've worked on since I developed semiconductor lasers to fiber optic communications.

VAMOSI: Id’ like to thank Michael Coden for talking about his new database operating system project, DBOS. If you’re at RSAC 2023, he’s presenting along with his colleague Micahel Stonebreaker on Thursday morning. If you’can’t make it to RSAC this year, then check out dbos-project.github.io for more information on this open source project. The concept makes sense, we’can’t keep using operating systems designed in the 1960s for the rest of the 21st century. Whether a database operating system is that next logical step, I’ll leave that to you. After all, we have to start building security in instead of bolting it on later. And redefining out concept of what we need for an operating system today and in the future, well that’s the first step.