High performance IP networking over the backplane with next generation protocol factoring

High-performance embedded systems often comprise multiple single-board computers (SBCs) connected via a common backplane bus. The SBCs usually communicate using the Internet Protocol (IP) over Ethernet, but using Ethernet requires that the system contain an Ethernet switch and, potentially, additional cabling. The logical alternative is to enable IP-based communication over the backplane, taking advantage of the backplane's performance potential. We propose high-performance, Linux-based, IP-compatible process-to-process communication using standard multicomputer backplanes for physical data transport. Because backplane accesses are privileged operations, the implementation must reside in the Linux kernel, and hence will consist of one or more loadable kernel modules. The software architecture will provide a framework for IP-compatible communication across general shared-memory interconnects. Distinct components within the framework will support specific standards such as Versa Module Eurocard (VME) and Compact PCI (cPCI), and will exploit optional performance-enhancing features of common devices (e.g., DMA). This structure will provide the flexibility to support both future backplane protocols and enhanced feature sets in future backplane adapter devices. Since the reliability of the communication protocol is critical to the stability of a distributed system, we also propose applying Reservoir's expertise with dynamic and formal verification to the implementation.