ParAccel Promotes "Dynamic" Analytics
Dynamic analytics describes an analytic use-model that emphasizes both discovery and iteration. The idea is to refine analytic discoveries by tweaking algorithms and queries.
- By Stephen Swoyer
- July 31, 2012
Don't talk to John Santaferraro, vice president of solutions and product marketing with analytic database specialist ParAccel Inc., about benchmarks. Neither Santaferraro nor ParAccel much want to discuss them these days.
It's not because ParAccel is a poor benchmark performer. Prior to last year's reorganization, in fact, ParAccel's marketing was dominated by benchmarking: its marketing collateral used ParAccel's performance in industry benchmarks to buttress its claim that it was one of the fastest DBMSes on the market. This was consistent with ParAccel's (founding) go-to-market strategy.
"Most of the marketing had looked at ParAccel as a very fast database, which it is," Santaferraro explains. "The goal of the company originally was to create the world's fastest analytic database. They've done a really good job of this. It's columnar. It has compression. We compile our queries so that they actually run in the database as part of the program. It was built from the ground up to do analytics," Santaferraro continues, explaining that much of ParAccel's recent development has focused on improving its manageability, connectivity, and analytic feature sets.
The upshot, he claims, is that the ParAccel of today is a proper platform for analytics: "[We've] repositioned [ParAccel] as more of an anlaytic platform as opposed to just a database."
That being said, ParAccel's former focus on benchmarking wasn't exactly a slam dunk. Competitors such as the former Sun Microsystems Inc. vigorously contested some of its claims, and a few vocal analysts likewise attacked its methods.
Its benchmark-centric approach to marketing also exposed ParAccel to predictable counter-messaging: sure, competitors used to say, it's won a few benchmarks -- chiefly because it either refuses to participate in or ultimately pulls out of any test it's likely to lose.
Five years after ParAccel's debut, and more than a decade after appliance specialist Netezza Inc. (now part of IBM Corp.) basically re-invented the category of the analytic database, benchmarks are passé. ParAccel's competitors, for example, have shifted their marketing to focus on issues other than performance. Santaferraro stresses that ParAccel's shift in emphasis wasn't an opportunistic (or marketing-led) one. Instead, he claims, it's backed up by solid development on the platform back end. "The questions that we asked ourselves were 'How do we extend the power of the database in ways to make it more of a platform for the analyst, and what does an analyst need to do to be effective?'" he explains.
The answer to the first question, Santaferraro and ParAccel determined, was a robust embedded analytic capability. "We embed 502 analytic functions right in the database. We do it in a library approach, so that the analyst can very easily using SQL call and run some of those analytic functions that sit in the library," he says. "That approach makes it very easy for other business analysts and SQL users to run those functions without having to be data scientists. They can use what's already in the library, so a MicroStrategy user or a[n SAP] BusinessObjects user could actually begin using more analytics than they were used to using with just those platforms."
This gets at perhaps the hardest problem in advanced analytics: how does one enable or equip non-expert users -- e.g., smart, capable, and Excel-savvy business analysts -- to interact effectively with advanced analytic technologies? Santaferraro and ParAccel say they have an answer. "We created the concept of on-demand integration. We created a series of on-demand integration models that allow the user to go out and get additional models at the point [where] they're running the query. They can get data from a database, they can pull in information from Hadoop, they can call Hadoop from a running job," he explains. "We have a series of these integration modules that are at different points of product development."
To kick things off, ParAccel developed an on-demand integration module for Teradata. It's likewise working on an ODBC on-demand module, says Santaferraro, and is also cooking up a similar offering for Hadoop. "Our Hadoop strategy currently is an on-demand integration module. It will go out and either pull in data from Hadoop at the point of running a query or collaborate and run a MapReduce job and pull in the result set."
Speaking of Hadoop (and, by implication, of big data), Santaferraro points to the emergence of a new class of big data-driven advanced analytics, which he calls "dynamic analytics." That's "dynamic" as in dynamically discoverable. Think of it as an analytic behavioral model that's uniquely suited to the way in which business analysts might prefer to operate.
"It's part of the analytic discovery process. The way an analyst does their work, they'll get some data, they'll run a [query] against it, [and] they'll see what [result] they get. When they tweak the algorithm, they get a different set of data," he explains. "The discovery process is this very iterative and dynamic kind of thing. Existing [i.e., conventional] data warehouse technology doesn't support large data sets [involving] complex queries, especially when you're trying to do ad hoc kinds of things."