In-Depth
Appliance Vendors Downplay Microsoft's Fast-Track DW Push
Appliance proponents aren't worried about SQL Server Fast Track Data Warehouse. Instead, they raise credible objections about Redmond's high-end DW push.
Last month, Microsoft Corp. announced SQL Server Fast Track Data Warehouse, a reference architecture program designed to predictably scale SQL Server 2008 to data warehouse (DW) configurations of up to 32 TB.
On its face, Fast Track could be a serious challenge to the data warehouse appliance players that have targeted SQL Server's shortcomings in large DW configurations, but DW appliance vendors aren't worried. Even those vendors that may be the most impacted by Microsoft's move -- such as Dataupia Inc. and ParAccel Inc. -- raise credible objections about Redmond's Fast Track program, starting with its lack of any massively parallel processing (MPP) component. DW appliance customers -- a handful of whom are running on top of SQL Server -- also raise concerns.
Customers who have taken the DW appliance plunge say the experience can be eye-opening. Take ITIS Holdings, a provider of both historical and real-time traffic and travel information, which implemented Dataupia’s Satori server appliance to accelerate its bottlenecked SQL Server 2005-based infrastructure.
“Our SQL Server [data warehouse] started off quite small, but things continued to get larger and larger. As our data sets grew bigger, with [ITIS’ users] running multiple queries, we started getting bottlenecks. We quickly decided that top-range solutions like Teradata were much more powerful ... and much more expensive than we required,” says John O’Reilly, database manager with ITIS.
ITIS had been using indexes to compensate for some of SQL Server’s performance shortcomings. This approach proved to be untenable in the long run, however. “Performance degradation was getting much, much worse as data increased. Our existing [SQL Server] systems relied very heavily on indexes, and the indexes were getting very large as our data sets grew. We needed to add indexes to get performance, but we didn’t have the disk space available to add them,” O’Reilly explains, adding that indexing with “Dataupia is much less of a requirement than with our traditional SQL Server system.”
Nor did ITIS want or need to jettison its SQL Server-based infrastructure. Apart from its substantial base of in-house SQL Server expertise, that database performed very well with OLTP-oriented applications.
“We have a lot of skills very focused around SQL Server, and [going with] Dataupia allowed us to keep those skills. We run traditional OLTP on SQL Server, too, and we weren’t having any problems there,” O’Reilly continues. “Where we wanted to look into using Dataupia was data analysis ... [for] live traffic information -- real-time traffic analysis. It was becoming quite difficult to get that kind of information out of SQL Server in a timely manner.” ITIS’ experience with Dataupia also illustrates one of the problems that Microsoft (and other traditional DBMS vendors) will have in luring back apostate DW appliance customers – or in convincing fence-sitting shops to stay the course: the undeniable performance advantage of MPP for analytic workloads.
Microsoft plans to deliver a native SQL Server MPP capability with its Project Madison effort, which is slated to ship next year; SQL Server Fast Track is not an MPP offering. Oracle offers MPP (or MPP-like) performance in its Database Machine and Exadata storage offerings, but -- at $60,000 or $130,000 per TB (per analytic DW expert Curt Monash’s latest figures) -- that offering is substantially more expensive than the $10,000-per-TB low-end figure touted by Dataupia (not to mention the $16,500-per-TB and $18,500-per-TB figures trumpeted by Teradata and Netezza, respectively).
ITIS’ O’Reilly says Dataupia’s Satori server is significantly faster than vanilla SQL Server -- and more scalable to boot. “In general we are finding a lot of the queries we are doing now run approximately four times faster than they were running on our legacy system. That performance is linear, so if we need to add capacity, we can just add another 2 TB blade and we’ll get a linear boost in performance,” he concludes.
Says John O'Brien, chief technology officer (CTO) with Dataupia, "What Microsoft is doing, it's the same thing as when Oracle announced their [Oracle Optimized Warehouse] reference architectures. I know all of the Oracle customers that really need that kind of information. They need the reference configurations to help them -- there's really nothing out there; consultants are the ones who provide all of that. So there's value in what they're doing but they don't solve the problem that we solve,"
Like ParAccel, Dataupia pushes a coexistence message: customers don't want to gut their existing DW infrastructures, he says; in fact, O'Brien concedes, there are DW applications for which SQL Server 2008 is well suited. With this in mind, he positions Dataupia as a high-octane complement to SQL Server and Oracle data warehouse systems.
"There are customers who invested a lot in their environment. They want it to stay there; they don't want to migrate to another platform and replace everything and rewrite their applications -- but they did have a serious data-volume problem to solve."
Like most of his colleagues, O'Brien cautiously positions Microsoft's move as a good thing for DW appliance players. What's more, he points out, Dataupia's claimed $10,000-per-TB price point for Satori Server comes in at $3,500 less (on a per-TB basis) than Microsoft's MSRP for Fast Track. "We're quietly optimistic. It does play to our strengths -- which was the low TCO around data management -- around on-demand scalability. You don't have to build out a bigger environment; you can add more as you need to grow," O'Brien indicates.
Industry veteran Kim Stanick, vice-president of marketing with ParAccel, has a similar take. ParAccel, like Dataupia, can be implemented either as a DW platform or (in its Amigo incarnation) as a complement to Oracle or SQL Server databases. What's more, and unlike Dataupia, ParAccel adds a columnar capability to its MPP underpinnings. In addition to pre-fab hardware and software bundles, ParAccel sells its software separately, Stanick notes -- bringing its price-per-TB down to less than $6,000.
Like other DW appliance proponents, Stanick uses a familiar figure of speech to describe the performance of DB2, Oracle, and SQL Server as data warehouse platforms. They're fine for smaller DW configurations (with limited users, and with a finite diversity of query types), she argues, but tend to "hit a wall" when scaled into double-digit terabyte territory. Their use in high-end, double-digit terabyte implementations is a function of two drivers, Stanick maintains: IT's comfort level with technologies it knows (namely, brand-name relational database platforms) and a one-size-fits-all philosophy which emphasizes platform standardization to help reduce administrative or other costs.
One-size-fits-all rarely makes sense in practice, Stanick contends. In the case of analytic databases -- where Teradata Corp., Netezza Inc., Dataupia, ParAccel, and other players can all trumpet anecdotes that have users of large Oracle or SQL Server data warehouses going from response times of several minutes down to the sub-second level -- a one-size-fits-all approach can leave companies at a huge disadvantage with respect to their competitors.
"Whereas it used to be that people wanted a one-size-fits-all solution, or they wanted a limited set of technologies, the whole mode of appliances has helped people realize that they don't have to be as limited with their selections anymore. It doesn't have to become a religious war anymore," she comments.
"I don't know if appliances caused this or if they're feeding the fire, but -- again -- the reason shops started to consolidate is to reduce the complexity and their overall costs, because of the human costs of having to manage a very mixed management environment. Once they did that, they found that [their data warehouse systems] were just hitting up against a wall: their SQL Server or Oracle data warehouses just aren't responsive enough for their business needs."
Tim Young, vice-president of corporate marketing for Netezza, favors a different figure of speech -- that of the leaky pipe. His message, however, is eerily familiar: DW appliances redress the shortcomings -- typically poor query performance (which Young says is best measured by breaches in service-level agreements, or SLAs) -- of brand name DBMS warehouses.
The point, Young says, isn't that Microsoft's SQL Server Fast Track configurations -- or Oracle's Optimized Warehouse (OOW) systems, for that matter -- can't or don't scale. It's rather an issue of their not being the best solutions for particular business requirements. In a tough economic climate, he argues, customers are going to think more tactically than strategically. In this context, the platform standardization or enterprise data warehouse (EDW) initiatives that once determined their strategies will get shifted to the backburner.
"We have always sold tactically. We sell a solution to a leaky pipe. A customer basically has a problem with a particular application -- revenue assurance, loyalty management, clickstream analysis -- and that problem shows up in terms of SLA breaches. We would come in and be considered as the solution for that particular problem," he says. "In a recession, that's all people think about. No one is thinking about corporate-wide, enterprise-wide, or culture-changing projects. Our view is that in most cases the advantage with the better performance will offset any perceived disadvantage [with regard to] standardization. In the vast majority of cases, the Oracle or SQL Server decision is based on convenience for the IT department."