Processing datastreams in real time

Prognosticators say that in the next 10 to 20 years, anything of value will be given a sensor tag so we can track its status and location, or both. And the network implications will be huge, with real-time streams of data flowing every which way.

One start-up formed to deal with the new demands is StreamBase Systems, which has developed what it calls a stream processing engine. The systems software (for Linux, Solaris and next month Windows) is designed to process real-time streaming data feeds in milliseconds.

The product is commercialization of research that has been three years in the making at MIT, Brown and Brandeis universities, says founder and CTO Michael Stonebraker. Stonebraker was the main architect of the Ingres relational database management system (DBMS).

There are two ways of dealing with messages in a stream, he says: examine and route them, which is fairly easy; or correlate the messages - even if they arrive at different times - and process them. That is harder, especially when the objective is to do it in milliseconds.

Financial service firms have applications that demand that capability, but Stonebraker sees future demand in everything from the military (data streaming from battlefield sensors), to industrial control and weather tracking.

For now, the company is focused on financial services, helping companies do things like rationalize information from multiple stock feeds. You can't do that fast enough using a relational DBMS because that requires writing the data, logging it, indexing it and committing the transaction, Stonebraker says. That's why financial firms tend to build their own systems at great time and expense.

StreamBase can process data on the fly using what it calls StreamSQL. It essentially adds windows (defined by time or by number of messages) that force SQL joins and aggregates without storing the data. And it can correlate information from multiple streams.

The result is a system that process stream data "a couple of orders of magnitude faster than the relational DBMS approach," Stonebraker says. And the tool's graphical development environment means customers can build applications in days instead of years, he says.

To win converts, the company tells would-be customers it will fix their hardest message stream problem in a week, even before talking sales.

StreamBase - which will officially launch at the Demo conference Feb. 13 - looks to have an interesting product. But we can expect a slew of companies to emerge over the coming years to deal with the explosion in sensor networks. This could be the next big thing.

Join the Network World communities on Facebook and LinkedIn to comment on topics that are top of mind.
Now read: Getting grounded in IoT