News

Analysis: Behind Microsoft's DATAllegro Acquisition

You've already snapped up Microsoft-branded mice, keyboards and gaming consoles. Is a Microsoft data warehouse (DW) appliance in your future?

• By Stephen Swoyer
• July 28, 2008

Just why Redmond grabbed DATAllegro is a matter of some debate. DATAllegro, like DW appliance pioneer Netezza Inc., was among the first wave of vendors to champion the turnkey DW appliance model. In addition, DATAllegro was arguably the first appliance player to shift production to all-commodity hardware from OEMs such as Dell Computer Corp., Cisco Systems Inc. and EMC Corp. (Rival Netezza still uses "commodity" PowerPC chips in its appliance architecture.)

Since then, a raft of players -- InfoBright Corp., ParAccel Inc. and Vertica Inc., among start-ups; Hewlett-Packard Co., IBM Corp., Oracle Corp. and Sybase Inc. among established vendors -- have also taken the DW appliance plunge.

Perhaps the driver was that DATAllegro gives Microsoft a symmetrical counter to relational database rivals IBM and Oracle on the DW appliance tip. After all, Big Blue has been shipping its DB2-based Balanced Warehouses (nee, Balanced Configuration Units) for four years now, while Oracle -- after years of informally marketing "reference" warehouse configurations that it developed in tandem with hardware partners such as HP and Sun Microsystems Inc. -- announced its first explicit DW appliance program, the Oracle Optimized Warehouse, last September. Perhaps the Redmond giant perceived a DW appliance gap: It could counter IBM's and Oracle's efforts in the data warehousing low-end (the sub- or single-TB range) but was running out of gas in the DW high-end.

DATAllegro CEO Stuart Frost, for his part, said that's precisely it. "[DATAllegro] is fundamentally a high-end play. It's fairly common knowledge that Microsoft's share of the market above 10 TB is much smaller than their share below that level," he said, stressing that "SQL Server has certainly improved greatly since SQL Server 2000, and SQL Server 2008 is a great SMP product. But that's all it is: SMP. Clearly we take [SMP] in a whole new dimension of scale-out."

Frost noted the companies' history of partnership: Redmond first approached DATAllegro with the suggestion that the two companies join forces to deliver an accelerated version of SQL Server. In the midst of that, Frost said, Microsoft and DATAllegro shifted gears, talking up a plan to develop reference architectures -- just as Oracle did with its OOW push -- for SQL Server in certain hardware configurations. Buy-out talks followed, he noted.

"Reference architectures are the way major vendors have co-opted the appliance story. A big part of why [Microsoft] bought us is that we have the expertise and the ability to take that [reference architecture] to a whole new level and leapfrog Oracle in [terms of] getting that reference architecture right," Frost said.

Were scalability and reference architectures the prime drivers? No one knows for sure -- but analysts are speculating. Consider veteran data warehouse architect Mark Madsen, a principal with consultancy Third Nature Inc. and author of several data warehousing books, who thinks the DATAllegro deal does, in fact, boil down to an issue of high-end scalability. DATAllegro, with its homegrown DW technology, had it. Microsoft, with its credible -- but, compared to offerings from Oracle and IBM, immature -- SQL Server database, didn't.

"It's the missing high-end scalability for SQL Server. [Microsoft] had nothing to compete with Oracle or IBM. Now they do," Madsen said. He also noted that Microsoft should have the goods to compete with Oracle and IBM "in three years, when the next rev of SQL Server comes out."

The good news, from Microsoft's perspective, is that DATAllegro -- unlike competitor Netezza, for example -- isn't tied to any specific hardware platform. It does depend on some special driver configurations, Madsen conceded, but not extensively. Furthermore, driver dependencies probably aren't anything that would prevent Microsoft from pushing its SQL Server-derived DATAllegro technology on to any of its existing partner platforms.

As for Madsen's timetable, Frost is far more sanguine. "It's not going to take years, as some people in the blogosphere are predicting," he said. "Just from [the integration work] we've already done, we've actually found that it's going to be pretty straightforward. All of the hooks are there already [such as] the APIs. We don't have to change a line of code in SQL Server."

He continued, "Yes, we'll have to change some of our query optimizations to make it run better with SQL Server. In most cases, however, it's that much easier [than working with Ingres]. SQL Server is just that much more sophisticated, [so things such as] join capabilities, I/O -- basically all of those things that we used to have to work around in Ingres -- we can just pass a lot of that through to SQL Server without optimizing it."

The more vexing question involves DATAllegro's software special sauce, which is based on Ingres and Linux, two technologies that are largely unpalatable -- if not anathema -- to Microsoft. "DATAllegro's code is on Ingres and Linux, but as I understand...[it], it's more layered on top than hooked deep inside like [rival appliance architectures such as] Greenplum or Netezza," Madsen said. "It certainly doesn't run on Windows or C#/.NET or on SQL Server."

That's true, too, Frost conceded, but DATAllegro is really more of an architectural vision that doesn't depend on any particular technology prescription. For this reason, he argued, the move from Ingres and Linux to SQL Server and Windows -- while far from non-trivial -- shouldn't be a showstopper.

"We took some decisions very early on, knowing that in such a large market dominated by an incumbent, [that our] most likely exit was going to be [by way of] an acquisition. So, for example, all of the communication between our end nodes [which take in data] is through standard SQL, so pretty much any database will speak the ANSI SQL that we use at that level. We do have some mechanisms to do sideways movements of data, but we've found that SQL Server with its APIs will allow us to port over pretty easily what we already have," he said.

Why Now?
As for the timing of the acquisition, Madsen, like other industry watchers, is flat-out flummoxed.

For one thing, he pointed out, DATAllegro recently closed the books on another round of venture capital (VC) funding, which likely boosted the price that Redmond had to pay for it. He stressed that the DW appliance segment is teeming. It's flush with vendors, ideas and -- to a degree -- with VC seed money.

That's bound to change: The market itself will contract, a natural culling will occur and VC funds will flow to vendors that have the best shot at making it. Why didn't Microsoft wait for this to happen, Madsen wondered. Why did it overpay now for what it possibly could have had -- six months from now, a year from now or 18 months out -- at a cheaper price? Madsen, like several other industry watchers with whom we spoke, doesn't know. "It's very not-Microsoft, so I don't get that part of it," he confessed.

There's another wrinkle here, too: Is a Microsoft-branded DW appliance in the works? Microsoft's past forays into hardware have been niche-y and limited -- largely to avoid upsetting its Wintel OEMs. The company has had considerable success with Microsoft-branded mice, keyboards and other accessories, for example. Does it have any real interest in marketing full-blown DW appliances, or is it just acquiring DATAllegro for its technology? Does it, in fact, plan to stop selling DATAllegro- or Microsoft-branded DW hardware?

Frost, with the obvious disclaimer that he isn't yet speaking for Microsoft, said yes. He explained that the DATAllegro buy signals a shift away from fixed-configuration appliances to reference architectures (similar to what Oracle is doing with its OOW initiative) -- a data warehouse vendor works with many hardware partners to build, test, optimize and certify its offerings in specific hardware configurations. To a degree, he suggested, that's where the appliance market itself is heading: Not just Oracle, but Sybase -- with its IBM System p-powered Analytic Appliance -- is another example in kind.

"The answer, really, is yes. It's pretty obvious, frankly. It makes a great deal of sense. Microsoft is not in the business of selling big iron. This is a lot of hardware, so why not partner with the existing vendors to deliver it?" he said.