Redundant hardware with hot-swap capability and system interconnects form the basis for high-availability hardware. Hot-swap support is key to continuing system operation because a fallback to redundant systems typically results in reduced performance or reduced redundancy. Replacing failed units restores a system's ability to weather subsequent failures.
A number of hardware technologies can be employed in high-availability embedded systems. Among others are CompactPCI, Ethernet, and InfiniBand. Operation of hot-swap subsystems, such as RAID tape and disk systems, are normally transparent to an RTOS.
The PCI Industrial Computer Manufacturers Group (PICMG) specifies CompactPCI's hardware characteristics, and the new PICMG 2.12 guideline includes hot-swap, an API that provides a consistent architecture for handling CompactPCI's hot-swap hardware.
The PICMG 2.12 API was developed by the Software Interoperability Technical Subcommittee chaired by Mark Overgaard of Pigeon Point Systems. The API can be incorporated into any RTOS, allowing applications to handle hot-swap CompactPCI card changes without dealing with vendor-specific APIs. The PICMG 2.12 includes OS-specific appendices for the Microsoft Windows NT family of operating systems, including Windows 2000 and Linux.
Ethernet has supported hot-swapping since its inception. An RTOS typically takes advantage of the network interconnect and is oblivious to how packets are moved between network nodes.
An RTOS normally utilizes heartbeat messages or more sophisticated communication methods to interact with applications distributed across nodes on the network. Cluster technology is a good example of this, although clustering support tends to be specific to an RTOS.
A variety of storage-oriented interconnect technologies have been employed in high-availability system designs. Among them are shared SCSI and Fibre Channel systems.
InfiniBand is poised for a move to the forefront of high-availability system designs. Its switch-based, mesh interconnect provides high-bandwidth connections between any hot-swapable node.
InfiniBand management software from companies like Lane 15 Software and Vieo offer a mechanism for applications and operating systems to track changes in switches and nodes.