It Takes Two: Microsoft's DW Market Strategy Explained

Microsoft sees a bifurcated data warehousing market, with Fast Track SQL Server on the low end and SQL Server Parallel Data Warehouse at the top.

When it comes to data warehousing (DW), Microsoft Corp. has several irons in the fire.

Actually, Microsoft has irons in a couple of different DW fires, There's Fast Track SQL Server, which it prescribes for DW configurations of up to about 20 TB. Fast Track officially launched in 2009, even though Microsoft and several hardware partners had been offering highly optimized SQL Server DW packages for several years prior to its debut. Competitive offerings from both IBM Corp. (its erstwhile DB2 "Balanced Configuration Units") and Oracle Corp. (its "Oracle Optimized Warehouse") couple SMP brawn with pre-configured and highly tuned (or "balanced") SQL Server, DB2, or Oracle 11g database configurations.

Recently, Microsoft stoked another fire: namely, its SQL Server 2008 Parallel Data Warehouse (PDW), a massively parallel processing (MPP) appliance offering -- based on technology acquired from the former DATAllegro Corp. -- that Redmond says can scale to support configurations of several hundred terabytes.

Microsoft sees a bifurcated DW market, with Fast Track on the low end and PDW at the top. True, there might be overlap at the fringes -- with some Fast Track implementations scaling into PDW territory, and vice versa -- but both offerings are designed to address specific (and, to a degree, mutually exclusive) market needs, Microsoft officials maintain. In other words, says Herain Oberoi, director of product management for SQL Server with Microsoft, PDW is no more suited for Fast Track-type implementation requirements than Fast Track is suited for PDW-type scenarios.

"Currently, we do want to continue having Fast Track and a software-specific offering as our core value in that [sub-20TB] segment," he says. "Because of performance requirements, because of scalability requirements, because of the fact that you can usually get [superior] price/performance at that size … you can achieve lower per-terabyte price points [with Fast Track]."

Oberoi says he doesn't see a scenario in which PDW scales down. If anything, he suggests, both Fast Track and PDW will tend to scale up in lock-step, with the improving performance of the former determining the low-end threshold of the latter: "As we continue to make improvements to the query processing engine … and [add] in-memory [capabilities] to the core engine … within the SMP architecture, we'll continue to see more benefits there."

Microsoft's (Determinedly) Bifurcated DW Strategy

Microsoft's acquisition of DATAllegro was a proactive move: it predated the first serious consolidation in the specialty DW arena by about two years. Redmond purchased DATAllegro in July of 2008; EMC Corp. acquired Greenplum Software Inc., kickstarting a wave of DW acquisitions, in June of 2010, but it took Microsoft almost two and a half years to productize its DATAllegro assets.

Redmond announced PDW at its Professional Association for SQL Server (PASS) conference in November of last year. By then, DW market-leader Teradata Corp. had acquired the former KickFire Inc., IBM had acquired DW appliance pioneer Netezza Inc., and EMC had acquired Greenplum. More acquisitions followed: Hewlett-Packard Co. (HP), one of Microsoft's PDW hardware partners, nabbed Vertica Inc. in February; Teradata struck again last month, acquiring Aster Data Systems Inc.

All of these vendors, along with a passel of specialty DW players, are gunning for a triple-digit-terabyte segment that market-watchers (following the lead of vendors such as the former Aster Data) have dubbed "Big Data."

There's Big Money to be had in Big Data -- that's one reason why storage giant EMC purchased GreenPlum -- but market-watchers say the DW market's "sweet spot" is still in the 0.5 to single-digit-TB range.

This is the segment Microsoft targets with Fast Track. It's a segment that other analytic database vendors aren't shy about targeting, too. Prior to its acquisition by Teradata, KickFire explicitly catered to 0.5 TB customers. Ingres Corp., with its VectorWise database, is also up front about competing in sub-TB configurations, in addition to larger implementations. (Like Fast Track, VectorWise is a non-MPP architecture. Unlike Fast Track, it vectors its operations to accelerate performance. There's also Exasol AG, an MPP specialist based in Nuremberg, Germany, that markets an analytic DBMS (EXASolution) that costs as little as $6,000/TB.

SQL Server PDW, for its part, clocks in at about $13,000 per TB. Although that's roughly twice Exasol's bargain-basement price per TB, it's nonetheless in line with (and in many cases cheaper than) competitive offerings. Oberoi declines to quote a per-TB price for Fast Track, saying only that "it's much less than that" of PDW. For this reason, he maintains, Fast Track is best suited for the low-end or "sweet spot" of the DW market.

In addition to its low price, Microsoft believes Fast Track has at least one other thing going for it: SQL Server's ubiquity. Chances are, you already have a SQL Server installation -- and, as a result, SQL Server experience -- somewhere in-house, so why wouldn't you standardize on a scalable and cost-competitive Fast Track system that -- as Oberoi puts it -- permits you to leverage the in-house database management and application development skills you've gone to great lengths to recruit and retain? However, if this is true of Fast Track, isn't it also true of PDW?

Oberoi basically concedes as much. "The underlying premise for bringing PDW to market is [that] we've seen a level of success in a high-end transactional database [with vanilla SQL Server], and when we looked at the price for massive systems on MPP, we basically said, 'We can bring that same price/performance value proposition [to high-end data warehousing],'" he says.

Nevertheless, Oberoi insists, there's natural market segmentation between Fast Track customers on the one hand and PDW customers on the other. "The simplest way to think about it is that both SQL Server Enterprise [Edition] and Fast Track [are based on] an SMP model with the intent to scale into the tens of terabytes. Depending on the types of data sets you have and the optimizations you have done, you can [scale these platforms] into the tens of terabytes," he explains. "When you really get into the high end, that's when you want to switch from an SMP to an MPP architecture. That's where SQL Server Parallel Data Warehouse comes in."

This is like comparing Apples and Microsofts, in part because it assumes that SMP performance can match MPP performance -- for all workloads, for all kinds of analysis -- up to a certain point. To put it another way: can a 5 TB Fast Track system match the performance of a 5 TB PDW configuration?

No, it can't, but sheer performance isn't the salient issue here, according to Oberoi -- price/performance is. It's the difference between "good enough" performance at a lower price point (with Fast Track) and excellent performance (PDW) at a higher price point.

This still doesn't explain why Microsoft seems to have categorically ruled out marketing PDW to single- and low-double-digit-TB customers. After all, its MPP competitors are scrambling to compete in this very segment.

If there's a market there, why not go after it? Oberoi doubles down on the price/performance argument. Lower per-TB price points, he maintains, are "just not a core focus for MPP."