New Teradata CEO pursues cloud-based architecture

Oliver Ratzesberger is the new president and CEO of data warehouse and business intelligence software vendor Teradata. His Jan. 14 appointment comes about as cloud-based architecture and AI technologies portend big changes in the way organizations do analytics. Ratzesberger previously was Teradata's COO; he succeeds Victor Lund, who will stay on as executive chairman of the board.

Starting in the 1990s, Teradata brought massively parallel processing (MPP) to address data management and analytics problems at major companies. For example, its systems have been central to core supply chain management applications at Walmart. Ratzesberger came to the company in 2013 after stints at eBay and Metascale. We spoke with him to get his view on this key data business player's path ahead, including the new Vantage analytics platform that it released last October.

In a way, Teradata is the father of data warehousing. There are pros and cons to that. What do you see as the value of having a lineage and, at the same time, the challenge of being innovative and new?

Oliver Ratzesberger: That is why I joined Teradata almost six years ago. Having been a customer and using the technology to solve some of the biggest data-driven problems at eBay, I felt that Teradata absolutely had the heritage for data warehousing and big data, but I felt that it had a very much underappreciated core of technology, as well as people.

Oliver Ratzesberger

On one side, I feel that heritage has given us a technology advantage when it comes to the largest problems. Hence our focus on what we call the megadata companies, where tens of thousands of employees need to go after thousands of virtual instances of data. What Teradata needed was to move toward the future of cloud, edge computing, microservices and APIs -- a large transformation was necessary.

There has been so much attention to Hadoop in recent years, sometimes as an alternative to the data warehouse. Is that lessening?

Ratzesberger: Well, you know, during my time at eBay, we were one of the very early adopters of Hadoop. We found it to be very promising, but ultimately realized that it was also overhyped and didn't deliver. And it forced us, then I was a customer, to go back to the basics and say, 'OK, we need to scale, we need workload management, we need to deal with complexity and high concurrency.' That seems to be foundational. And we've seen this in our customer base.

Hadoop has actually been on a retreat quite a bit. There are still instances of Hadoop -- there are still data lakes out there. The whole Hadoop ecosystem is shifting because it is quickly being kind of displaced, or almost rewritten from an architectural perspective with cloud architecture.

That leads to the question -- how is Teradata adapting to cloud-based architecture?

Ratzesberger: We have done a lot of that rearchitecture over five years, making Teradata software that runs natively on the various cloud platforms, whether it's AWS or Azure or in private cloud environments.

Where we are heading is leveraging, for example, storage options -- object stores, S3, Blob Storage in Azure -- making these native storage options for Teradata to bring the various different storage formats together.

These are the kinds of storage pools that are quickly replacing other file systems that have been out there. And our customers see a great need for having that directly integrated in a platform like Vantage, rather than having to stand up separate clusters to deal with that.

We're just rolling that out. There's a whole roadmap for 2019. We currently support S3 through various mechanisms and there's more integration with that coming out later this year for those very reasons.

How do you judge the company's experience with Aster Data, the MPP software for big data analytics that Teradata bought in 2011?

Ratzesberger: When I joined six years ago, the question became, 'Why do we have Teradata and a separate platform called Aster?' None of our customers want separate platforms for these kinds of capabilities. What we have done over the last three years is we have completely integrated Aster into Teradata, and that product is Vantage now. That's the analytics platform. It has all the capabilities of Teradata and overlaid to that, all the analytical functions that Aster came up with over the years -- without the need for separate data stores, without the need for separate SQL engines -- as a single, integrated solution.

The data warehouse started life as a reporting mechanism and that's changed. How do you view machine learning and AI analytics in the context of the data warehouse? It seems to be the next step in a way.

Ratzesberger: We believe this is a very complementary kind of next step because machine learning and AI requires highly structured data. In fact, we need to structure data more than we've ever done before because AI algorithms and machine learning algorithms are best trained on structured aspects of data or in structured features and variables.

AI algorithms are very susceptible to bad data and can quickly deteriorate.

And that requires a lot of processing, preprocessing and feature extraction, in order to power AI algorithms. And this is what we can excel at. Plus, what AI algorithms really need is a very constant level of data quality.

AI algorithms are very susceptible to bad data and can quickly deteriorate in their effectiveness when the data quality is not under control, and that's something that a lot of companies around the world have now learned. Even in data warehousing, there was the saying 'bad data in, bad data out.'

In AI, it's even more the case, because once the algorithm starts making incorrect decisions based on bad data, the implications are even bigger. That is because you're operating at a much different scale.