Data Services Journal

If you’ve got simply scads of data – and why wouldn’t you? – it’s doubling every 18 months – and are shuttling it to an application for analysis, you’re doing it wrong.

That’s so…so, well, 1980.

According to Aster Data, applications need to go to “Big Data,” not the other way around.

And to do that the company’s got a massively parallel data-application server that can embed applications inside a massively scalable MPP data warehouse and analyze petabytes of data – or terabytes, if that’s all you’ve got – ultra-fast.

Apps are automatically parallelized for scale; users can take their existing Java, C, C++, C#, .NET, Perl and Python applications, MapReduce-enable them and push them down into the data.

The widgetry runs on a cluster of commodity boxes. Figure five servers to start although parallelized applications can utilize terabytes of memory and thousands of CPU cores.

This is not the data warehouses, DBMSes and data analytics solutions of the last three decades that have separated data from applications, a technique Aster says results in massive data movement, latency and restricted analysis.

Traditional systems weren’t built to process billions of rows of data in seconds or handle chi-chi stuff like real-time fraud detection, customer behavior modeling, merchandising optimization, affinity marketing, trending and simulations, trading surveillance and customer calling patterns.

They were built for data sampling, an inexact science. They simply fail in today’s big data, analytics-intensive environments, Aster says.

The company’s Aster Data 4.0 brings data and applications together in one system, fully parallelizing both, to deliver ultra-fast analysis on massive data scales. And it’s got customers like comScore, Full Tilt Poker, Telefonica I+D, SAS and MySpace, with the big clutch of data of all, saying it’s right.

Aster’s Massively Parallel Data-Application Server 4.0, based on research done at Stanford University before commercialization started a couple of years ago, lets companies embed application logic in Aster’s MPP database, which includes MapReduce. It was Aster that brought MapReduce to SQL, a trick it’s now building on.

In Aster’s system data management lives independent of the application processing but – and this is important – the data and applications execute as first-class citizens, with their own respective data and application management services.

The Data-Application Server is responsible for managing and coordinating the cluster’s activities and resource sharing. It acts as a host for the application processing and the data managed inside the cluster.

As a data host, it manages incremental scaling, fault tolerance and heterogeneous hardware for application processing and it manages workloads via Aster’s new Dynamic Workload Management (WLM) capability.

Aster says WLM, described as the first dynamic workload management capability available on a MPP system to run on commodity hardware, can support hundreds of concurrent mixed workloads. It manages data storage, transactional correctness, online backups and information lifecycles (ILM).

The separation of data management and application processing is supposed to provide maximum application portability so a wide range of applications can be pushed down into the system.

Aster says this data analysis architecture distinguishes its solution from lightweight implementations of MapReduce, including what some vendors refer to as ‘In-Database MapReduce.

Richard Zwicky, president of Enquisite, the company that provides search optimization software and solutions, says that with Aster Data, response times for large queries has dropped from five minutes to five-10 seconds, and queries that previously weren’t possible now can be executed in 20-30 seconds.

Aster Data is backed by Sequoia Capital, Jafco Ventures, IVP and Cambrian Ventures, as well as Google’s first investor David Cheriton, Ron Conway and Rajeev Motwani.