What’s Tableau doing buying HyPer, the German OLTP/OLAP hybrid? The March 10 announcement brought a flurry of email — and the Doodler notices news about a company he’s watched since it was a mere puppy that tipped over an industry. Now as it eases into midlife, flush with cash and renown, it does what comes with that age: It buys stuff.
Innovative, high performance database system
The newly acquired technology is, according to the Tableau announcement, an “innovative, high performance database system.” Like the Stanford University-developed Tableau, this database was also developed at a university, the Technical University of Munich. There, Tableau will add “key technical personnel” and establish a research and development center for high performance computing.
I checked first with Dan Graham, director of technical marketing at Teradata, who speculated based on Tableau’s scanty clues.
What could they be thinking? Hmmm. The press release only mentions “OLTP and OLAP.” Gartner calls the use of OLTP data in real time for analytics “HTAP” — hybrid transactional/analytical processing. So this leads us to guess that Tableau would like to analyze data in actual real time. This would certainly set them apart from competitors. This also suggests Tableau needs a solid real time data ingest sub-system to get the data from OLTP systems (Apache Kafka??). This makes them the BI counterpart to SAP HANA. HANA’s main theme is HTAP — do it all in one corpus of data.
While this is hugely attractive on a theoretical level, it is also incredibly difficult to do. Transactions can stall and remain “open” for long periods of time. Transactions are not worthy of analysis until they have been committed to persistent storage. Transaction schemas and analytic schemas are often in conflict regarding performance. And transaction data is not clean when it arrives “in-memory,” so it’s like Hadoop’s schema-on-read. Who cleans that data and when? Minor cleansing can be done in real time. But the cleansing and transformations applied to a data warehouse cannot be done in real time. Nevertheless, I would place a bet on the only clue Tableau gave us: HTAP BI. Hence, 10 percent of the TUM research could be made into algorithms inside Tableau.
Not a product
A second source sounded more skeptical. He said of the University of Munich’s research team, “These guys are academics. They’ve never had the 2 a.m. phone call.” The HyPer database “is not a product… I see no proof it’s been installed on an actual customer site.”
He said he expects that the Tableau product will add HTAP query processing — and that that will set off a “war” with other BI vendors.
Tableau chief development officer Chris Stolte is, of course, optimistic. He talks of the “wealth of talent from [the university].” That is, we just have to wait. They must see more in Munich than the beer.