Parallel computing is the way of the future. Compute clusters on the macro level and multiprocessors on the micro level are ganging up to accelerate the processing of huge quantities of data. Many advantages can be reaped if Internet Protocol (IP) networking can be taken to the tactical edge—where sensors collect valuable data. Why? Because IP networking is pervasive and inexpensive. And better than any other protocol we have today, IP networking can minimize the interoperability risks of developing applications on compute clusters in the laboratory and then deploying them to large systems in the field.
Why hasn’t this happened already? Mainly it’s because the IP protocol, while flexible, doesn’t provide the necessary real-time, deterministic, high-bandwidth processing capabilities that an embedded interconnect like RapidIO can offer. Therein lies the challenge for bringing IP networking to the tactical edge, where multiple sensors generate explosive signal processing requirements that must be answered at lightning speed.
Sensor Gateway Solutions
To address these challenges, a bridge in the form of a sensor gateway to interface among various protocols is needed. The gateway bridges a RapidIO subsystem with an external 10 Gigabit Ethernet network. It also forwards Ethernet traffic into and out of the embedded processing subsystem, which looks to Ethernet like a standard IP-based cluster. The processing subsystem can be integrated into a platform’s IP-based enterprise network and participate in supporting a more agile, flexible, and robust sensor computing and networking environment.
IP-based command and control packets pass through the gateway, making the subsystem and all its processing assets visible to the network and interoperable with all network-based assets inside and outside the subsystem. While external data traffic also passes through the gateway, internal data traffic, often more critical to sensor capabilities, is mapped directly onto direct memory access (DMA) channels, bypassing all the header processing associated with IP networking. The result is “converged sensor networking” that uses a sensor gateway to bridge traditional IP networking with high-performance sensor computing, based on a novel Converged Sensor Network (CSN) Architecture.
IP-Based Networking Benefits
Introducing IP-based networking into a signal processing subsystem has several benefits. One of these is that, compared to an embedded interconnect, an IP-based network has few if any location constraints. By bringing the network deep inside a processing platform, application functions can reside wherever is optimal for performance, productivity, and life-cycle considerations, even outside the boundaries of the platform. Another benefit is the low cost of ownership of IP-based networks. Unlike almost any other technology, interoperability risks, from the laboratory all the way through multiple technology insertions in the field, are largely transparent to those who adopt IP as a transport fabric for their application.
Many systems today provide access to key information on a system-by-system basis. The user of these systems is, in effect, the integrator. Another benefit of IP-based networking is that it takes over the integration problem, so the user can obtain broader information from a net-centric vantage point, essentially moving from point-to-point interactions to many-to-many interactions. To efficiently exploit a range of sophisticated sensors for a variety of application needs, the CSN Architecture supports not only the computing required to transform raw data into information, but also the networking and exploitation capabilities to collaborate among sensors on the collection of integrated data that directly addresses the user’s requests.
The increased flexibility and ease of integration inherent in the CSN Architecture enable many usage models for signal processing resources with deeply embedded network capabilities. One such usage model is sensor networking, where sensors transmit raw data in the form of IP packet streams. CSN exposes the signal processing assets to network management for command, control, and high-availability management. In addition to IP data streams injected into the RapidIO subsystem, applications can rely on network connectivity to facilitate and comply with various management-level protocols native to standard Ethernet.