Andy Mendelsohn:
Oracle and The Autonomous Database

• Introduction
• Autonomous Database in Historical Context
• The "Self-Driving" Database
• Autonomous Database vs. Traditional Database
• Technical Challenges Building the Autonomous Database
• Business Benefits of the Oracle Autonomous Database
• Advice for CIOs

This transcript has been lightly edited.

Introduction

Michael Krigsman: We're talking with Andy Mendelsohn, who is the executive vice president responsible for the database at Oracle. I'm Michael Krigsman. I'm an industry analyst and the host of CXOTalk. Andy, thank you so much for taking time to be with us today.

Andy, you have been at Oracle for many years and so give us some historical context of where the autonomous database fits into this long Oracle history.

Autonomous Database in Historical Context

Andy Mendelsohn: Oracle, as a company, goes way back to, actually, the late '70s or early '80s. We started out as a software company focusing on what was then a new area, relational database management. If you think about the history of technology, there are very few technology companies from that era that are still left.

I think one of the remarkable things about Oracle and the Oracle database technology we build is how we've been very good at making sure this technology is the market-leading database technology, as we go from one generation to the next. In the '80s, it was the mainframes. In the '90s, the Internet came along. Now we're moving to cloud.

What we've done really well at Oracle is we've navigated these changes from one generation to the next. That's what autonomous database is all about. We're now navigating this change from the Internet era to the cloud era, from the software side, and making sure, again, we have the best database technology for the next generation, which is what we embody into this whole autonomous database concept.

The "Self-Driving" Database

Michael Krigsman: What has been the key, if you can put your finger on something that has enabled you to remain relevant to such a profound degree through these changes in technology over such a long time?

Andy Mendelsohn: The big thing we had done was, we figured out how to make the Oracle database scale out for all sorts of workloads, especially transaction processing workloads. To this day, no other relational database vendor has ever figured out how to do that. We have this technology called Real Application Clusters, which is very unique. None of the other vendors were able to copy it, and so it gave us this huge differentiator as we went to the internet era.

As we're going now into the cloud era, what we think customers want is they want to eliminate all the friction around running Oracle databases, relational databases, all the friction around developing applications around databases. What customers really want now is they want to focus on their business. They want to focus on, "I'm a bank. I want to focus on running my bank better. I want to understand how to better market to my customers."

They don't want to run IT anymore, and so the big thing we're working on is eliminating all the human labor around using provisioning, running, operating, building applications around the Oracle database and just let people do what they want. They want to run their business. They want to leverage the value out of all that data they have.

At a high level, what we are doing is we are creating a self-driving database. Everybody sort of understands what a self-driving car is. You go into the car, you plug in the address, and the car is supposed to take you there. We want to do the same thing. We want our customers or our developers to be able to just say, "Okay, the data engineers say this is the data we want to model. Create a set of tables for storing that information in the database."

Then, after that, you just let your organization loose on it. Your developers can now just build their applications against that data. They can load up the data. They can run some SQL queries.

We will do everything for them. We will provision the infrastructure for them under that database. We will make it highly available for them. We'll give them an SLA if they want. We will make it fully secure, implementing all of the best practices for them. We take over all the tedious efforts that they have in their traditional organization for provisioning databases, for upgrading databases, for patching databases, for tuning the databases.

Autonomous Database vs. Traditional Database

Michael Krigsman: Andy, to place this into greater context, explain how this is different from traditional databases.

Andy Mendelsohn: At the foundation, with an autonomous database, we are using a very optimized infrastructure for running the Oracle database that's based on our Exadata technology. That is now a very mature platform and, in our cloud data centers our in cloud at a customer where we run the customer datacenter, we base our infrastructure on this Exadata technology, which is an optimized platform of hardware, servers, storage, networking technologies, and the database software that gives customers a very high performance, highly reliable, highly secure platform for running your Oracle database.

That's the foundation, a very mature infrastructure technology that we developed originally for on-prem and has now become sort of cloud-native technology where you can provision the equivalent of an Exadata box with the push of a button. Then we're implementing best practices around how you make database secure, how you make them highly available. That's sort of the database technology core.

In the database space, we implement. We are leveraging an autonomous database. A lot of this differentiated technology we've built over the years. Beyond the Exadata technology, we have the rack clustering technology I mentioned earlier that we built many years ago for doing scale out of Oracle databases, so we have a very powerful, shareable infrastructure for the database.

On top of that, we're using our multi-tenant technology. Multi-tenant means that this powerful infrastructure can be easily shared among modeable businesses. We create the equivalent of a virtual database that each customer gets when they ask to provision a database, an autonomous database. We don't give them the whole infrastructure. Essentially, consider it like a virtual database out of this large infrastructure.

Beyond that, we're taking care of the self-tuning of the customer's workload. At a high level, we're looking at two different kinds of workloads. Customers are doing analytics with their databases or they're doing more like run your business applications on the database, more operational workloads like running a business application like Fusion applications to do your accounting, e-commerce work, supply chain management, manufacturing, et cetera.

We have this common technology stack and we optimize it for these two kinds of workloads. There's an autonomous database for analytics. We call that autonomous data warehouse. There are autonomous databases for more operational use cases. We call that autonomous transaction processing. Then we have our self-tuning technology that goes into each of these two cases to give customers automatic, high performing systems.

Technical Challenges Building the Autonomous Database

Michael Krigsman: What were the kind of challenges or technical hurdles that you had to overcome to build the autonomous database?

Andy Mendelsohn: If you think about what we're doing, our goal is to eliminate all the human labor that our customers either on the operations side or their developers or analysts side had to do around the database. This is a huge exercise in artificial intelligence. At each of the layers of the stack, we had to go through and figure out where is the human labor still present. We had to build an appropriate algorithm of some sort to automate what the humans do in that layer of the stack.

Exadata, I mentioned earlier, is a very mature platform we've been working on for many years since 2008. We've already pretty much automated all of the security and availability areas, and performance and scalability areas in that technology stack. We also have lots of technology that actually observed the hardware, predicts failures are going to happen, alerts people to sort of know that this hardware device is going to fail and we should replace it before it fails.

The new things we're working on there are understanding, when something fails, diagnosing what is wrong and then fixing it. This is one of the main things people do around databases and protection IT shops. We are working on building algorithms that basically track all of the unusual, exceptional events going on in the database.

We're tracking all the telemetry, all the statistics around performance related statistics coming out of the database. We're building machine learning algorithms to observe all this data and try to predict failures. If there is a failure, try to repair automatically from the failure. These are very difficult problems to solve and our engineers are very excited about working on them.

Michael Krigsman: Can you identify some particular aspect of the technology that was particularly difficult?

Andy Mendelsohn: One of the big technology, key technologies in these databases are things called query optimizers. Query optimizers have been around since the beginning of relational database technology in the '80s. These pieces of technology were early expert systems that were designed to look at customer SQL statements and figure out the best way of executing them for the highest performance, et cetera.

These algorithms, these optimizers are dependent on knowing the statistical distributions of the data in the database. As you load lots of data, the distributions change. For many years, one of the big problems customers had is, the query optimizer would choose the wrong way of executing a query because it didn't have up-to-date statistics on what the data looks like in a database.

One of the key things we did with autonomous databases is we said, "Okay. We're just going to go to real-time statistics," and what that means is that whenever a customer loads some data or they update some data, we're making sure all the statistical information that we use in the query optimizer is up-to-date with the actual data in the database. That makes the query optimizer a whole lot better at generating the right execution plans versus the way things used to work where the customers were responsible for periodically updating statistics when they sort of felt like it. That was one of the key things we did to make this all work really well.

Beyond that, one of the other things we've done is we've built an expert system around the query optimizer that really allows us to understand, be the super DBA. We can understand exactly what the customer workload is. We can do experimentation to figure out, "Oh, the workload has changed. Maybe we need a new index. Let's go off on the side, do some experiments, figure out what the best new index to use is for these queries. Put them into production."

We're doing this all online without any customers knowing we're doing this. The next time the customer runs a SQL statement that maybe can benefit from this new index, we will go use it. We know what the runtime for that query used to be. With the new index, of course, it should be faster and 99.99% of the time it will be. But if occasionally it's actually slower, we notice that, too, and we will pull that new plan out of production. This is one of the big benefits of an autonomous database versus what customers have today is we never have performance regression.

Business Benefits of the Oracle Autonomous Database

Michael Krigsman: From the business standpoint, is the primary benefit this labor efficiency and the fact that there are all of these database tasks that do not require labor anymore?

Andy Mendelsohn: Yes. If you're a CIO, you care about making sure the business is running well, at low risk and low cost. This autonomous database does this for them. They get this very high-performance technology that can run in a cloud. On a cloud, you pay per use, and so to the extent [that] we have much better performance than our competitors, we actually have lower costs.

There's a really radical change going on here where we now, in a cloud service, can offer data management that's the best in the world at the lowest cost. For a CIO, it's also really low risk. All these applications that have been written over the year, all the packaged off-the-shelf applications he's using, almost all of them use Oracle databases. We give them a very comfortable, low-risk way to go on a journey to the cloud.

The CTO types and developers, they love this because now they can do their job without any other organization in their way slowing them down. They can go straight to the cloud. They can provision databases. That's all they need to do. We take care of all the upgrades for them. We take care of the patching for them. It's a really nice world.

Michael Krigsman: It sounds like you're stripping out the low-value tasks and leaving the higher value activities.

Andy Mendelsohn: Yeah, that's exactly right. The way the business will look at this is that these people who they used to think of as DBAs are going to be much more valuable. They're data engineers now. They're data architectures. They're helping my analysts.

Yeah, exactly. They're going to move to a higher value role in the business.

Michael Krigsman: Andy, any thoughts on the general implications of these types of autonomous systems on organizations and on the future of work?

Andy Mendelsohn: Yeah, so autonomous systems are not just about a database. I'd mentioned at the beginning there are autonomous cars. I think it's all about empowering people.

Today, if you look at a person, they all have this phone they're carrying around that's doing all kinds of things that people used to have to do by themselves with 20 other devices or no devices at all. They had to do it by writing things down on a piece of paper. I think autonomous databases and autonomous systems, in general, are all about making humans much more empowered, much more powerful by giving us really powerful tools to do our jobs much more productively and do things we could never have done before in the past.

I think there's a bright future for autonomous systems. You just have to view them as systems that are helping empower people as opposed to systems that are replacing people. It's just really making people much more powerful and much more productive.

Advice for CIOs

Michael Krigsman: As we finish up, any thoughts or advice for CIOs who are looking at this and saying, "Yeah, I need to embrace this new world. I'm not sure how to do it"?

Andy Mendelsohn: Yeah, I think everybody is recognizing now that cloud is the next generation of computing technology and it's inevitable. Whether it happens in your organization in the next 2 years, the next 10 years, or 20 years, you have to start plotting out your roadmap, how you're going to get from where you are today to cloud.

One of the things that I see a lot of enterprise customers doing is they've invested a lot in building out applications over the last 25 years that are running in their data centers. They're not going to throw those things out, and they're not going to just lift and shift them magically to a cloud tomorrow. What they would like to do is transition, in an orderly way from where they are today, to a future where maybe they have no data center.

Michael Krigsman: Andy Mendelsohn of Oracle, thank you very, very much for taking time with us today.

Andy Mendelsohn: Okay. You're welcome.