Actian’s Ahmed Ezzat: Building Virtual Data Warehouses
Big Data needs fast virtual data warehouses, says Actian’s CTO Ahmed Ezzat
Big data firm Actian has a lot of heritage (it used to be called Ingres) and it has just acquired a bit more with the arrival of chief technologist Ahmed Ezzat.
Ezzat’s career started at Bell Labs, as one of a small team writing Unix 5.3 – a 1980s version of the operating system which inspired Linux and fed into commercial Unix operating systems including IBM’s AIX, HP’s HP-UX and Sun’s Solaris.
Since then he’s pioneered parallel processing at Cray Research and was Oracle’s lead for Java at a critical stage for that technology. Most recently he’s been working on Big Data analytics, as a distinguished Fellow in HP’s cloud division.
At Actian, he joins a company with one of the longest-standing database products and a self-proclaimed mission to bring big data analytics to new users.
An eventful career
Ezzat’s role as CTO is to help create the back end for Actian’s big data technology, the company has told me. I asked him how that fit with the rest of his career when he visited London.
The answer took me through a history of the last 25 years of the industry. At Bell Labs, Ezzat worked with Dennis Ritchie, who first created Unix, before shifting to Hewlett-Packard to work on compilers – it was “closer to development” than Bell Labs’ fundamental research, he explained.
He then hopped to Cray Research to be chief architect for T3D, Cray’s first parallel processing supercomputer. He enjoyed that but, “the weather was harsh in Minnesota. We had to move between buildings in underground tunnels,” he said.
He subsequently returned to the Bay area to work at Oracle, making its software run on symmetric multiprocessing (SMP) machines that were being sold for reliability by vendors including Tandem and Pyramid. “Pyramid was trying to work on 12 CPUs, but Oracle used to run on only single CPUs,” he said. “Oracle had a hard time scaling up to run on SMP.”
It was Ezzat’s job to design and implement the solution: “It took me a year or so.”
After that, he led Oracle’s Java efforts, at a time when Sun Microsystems was pushing Java very hard as a cross-platform application system. “I represented Oracle in the interface with Sun, in discussions on things like J2EE and JDBC 1.0.”
He then went to a start-up, Velosel, which was bought by Tibco, and then moved to fault-tolerant computer maker Tandem, which was eventually acquired by HP. Back at HP, he got stuck into some more innovation. “We moved SQL into storage, and HP sold Oracle on the idea,” he said. “It’s now known as Exadata.”
By the end of his stay at HP, Ezzat was working exclusively on analytics within the cloud, and along came a merger with Autonomy, which HP bought in 2011 for $10.3 billion. Was he not interested in the expansion of analytics within HP? Or are the stories true that the merger was not going at all well?
Perhaps unsurprisingly, Ezzat won’t talk about it, and he’s also not keen to explore whether Oracle’s current tactic of trying to pillage Java’s patent values could have been foreseen. This talk is supposed to be about Actian, after all.
Where the Actian is?
Actian is formed around the venerable Ingres database business, which was bought by CA and then spun out as an open source product, OpenIngres. The company became exciting to Ezzat a year or two back when it launched a vector database, VectorWise, which it claimed could make data analysis go ten times faster, cutting the cost of analytics and democratising it.
Last year, the company changed its name, shifting form the open source model, becoming Actian, taking on a new CEO (Steve Shine formerly of Sybase) and promoting “Action Apps” (which sound like dolls to us).
“This has the potential to do the things I really wanted to do all along,” said Ezzat. He also wants to work in a smaller company where the speed of execution is faster. There is still a good flow of revenue from the old Ingres business, and Ezzat “when I ask for resources, I get the resources and head count I need.”
SQL data is very well understood, he said, but things are changing. “The world is moving towards information management not data management.” The future is to do analytics on top of environments such as Hadoop which include unstructured data, he said: “There are very difficult issues in that area, which nobody has solved. I think I have some ideas about how to address most of them.”
SQL data isn’t going to go away, he said: “Vendors will differentiate themselves on how much they leverage the unstructured data.”
The critical thing for Actian, he said, is to integrate its technologies so it doesn’t create more silos. He wants to let businesses see data as a “virtual data warehouse”, where one API can query all the relevant data and leave the app to handle the business logic.
“If you want to deliver a lot of power, with little effort, our data platform will do that for you,” he said. “The guy who is going to develop the Action App does not have to be a professional programmer. There is a GUI tool and most of the code will be generated for you.”
There are two difference from previous attempts to work with data warehouses, he says. Firstly Actian doesn’t require an actual data warehouse to be built, with all the data extracted transformed and loaded into it (the ETL process), and secondly VectorWise allows processing that data to happen in real time, allowing predictive analytics.
“SAS, MicroStrategy and others can access different data types, but they use ETL which can never do that in real time for you,” he said. They also tend to rely on User Defined Functions (UDFs) which extend SQL.
“We will extend UDF to new types of UDF for new analytics functions, giving uniform access to heterogeneous data types,” he added.
In some ways, the job takes him back to what he does best as Actian has an issue he has seen before. Just as Oracle needed to make its software run on multiple CPUs, Actian has to scale VectorWise beyond one node.
He said that when that’s fixed, “VectorWise is going to be even more powerful than it is now. We are working hard to scale it out beyond one node. It is hard for someone with my background to resist that challenge.”
Ahmed Ezzat has been a Hewlett Packard Fellow and as a chief technologist in HP’s cloud division, he led the company’s work in analytics. As a senior director and architect at Oracle he worked on Java implementation. At Cray Research he managed the operating system group in the flagship T3D NUMA supercomputer product. He has served as Adjunct Professor at Santa Clara University’s Computer Science and Engineering Department for the past 20 years.
