Dawn Of The App Aware Network

OVER THE EDGE
At least three startups are also chasing the idea of non-XML application network appliances, though they're not aimed at the data center. Augusta Systems, Blue Vector Systems, and Omnitrol focus on processing data gathered from what they call "edge assets." Those include RFID readers and other sensors that are just now being linked into enterprise IT systems. The appliances aim to handle data from these assets locally, without involving a data center or a server.

"The network management issues that arise from the deluge of edge asset data require an intelligent network infrastructure," says Patrick Esposito, Augusta's president and chief operating officer. Such assets tend to be located in warehouses, factories, and other areas far from corporate data centers, giving this kind of appliance the same business case as WAN optimization--to avoid sending data over costly, slow WAN links.

Augusta started as a software company, selling .Net-based middleware called EdgeFrontier. Though most customers still install EdgeFrontier on Windows servers, Augusta expects to move into switches and routers; it's talking with at least two potential partners.

Omnitrol and Blue Vector also make appliances, again designed to be installed at remote locations. Omnitrol's includes a wireless switch that works with access points from 3Com and D-Link, aiming to take care of all of a branch office's communications needs. Blue Vector is more tightly focused on RFID, often combining its products with RFID readers. For example, it has an RFID-equipped refrigerator aimed at the pharmaceutical industry that automatically reorders drugs when supplies get low. It has also partnered with Nortel to use wireless mesh networks to link sensors.

Allegheny Power is using the Augusta boxes to build what it calls a self-healing electric grid. The SensorPorts process measurements from voltage sensors on power lines, temperature sensors in transformers, and moisture sensors in areas prone to flooding, rerouting power in case of an outage. "It's a way to transfer loads back and forth without having to rebuild our entire utility system," says Harley Mayfield, a planning engineer at Allegheny.

Sending meter readings all the way to a data center is not an option because the network is often out along with the power. The project is still just a pilot using a single SensorPort, but Allegheny plans to install 12 SensorPorts at substations, linked wirelessly to about 1,000 sensors throughout the grid. It also hopes to use the same infrastructure for automated meter readings for 1.5 million customers in Maryland and surrounding states.

Mayfield likes the way the appliance supports almost every type of sensor. "It's like tying a thermostat to a game console to a VCR," he says. "Being an old Star Trek fan, I look at the SensorPort as a universal translator." But as with the 3Com box, actually programming it isn't easy. He's had to bring in an outside company to provide custom software.

Still, that's an improvement over how sensor networks used to work. Developers working on sensor networks are used to debugging with an oscilloscope, says Joe Polastre, CTO of sensor software startup Sentilla. The company has ported a Java runtime to the 8-bit processors used in wireless motes, the dime-size nodes that make up wireless sensor networks, and sells an Eclipse-based SDK that lets developers cut and paste code directly from the Java apps used in most SOAs. Sentilla has signed up customers in the agricultural and border-security sectors. The business case is the same as with appliances and WAN optimizers writ small: to save bandwidth.

"The more processing you do on the sensor, the less you need a network at all," Polastre says. And with a tiny battery-powered device that consumes energy every time it transmits, avoiding the network means big savings in maintenance.

Sentilla and similar companies like Arch Rock could be the Achilles' heel of the movement toward intelligent networks. Once applications bust out of the data center, why should they stay in the network? Why not migrate all the way to endpoints, leaving the network as, again, just a pipe? Something to consider when deciding whether to buy into this new network vision.