OpenVPX/InfiniBand: From SWaP to SWaP-C

As the search intensifies for equipment with advantages in Size, Weight, and Power (SWaP), the military services are also focusing on system cost – up front and over the life cycle. Thus, like it or not, SWaP is becoming “SWaP-C.” Faced with budget cuts almost everywhere they look, service leaders are stressing affordability as never before.

One of the most attractive and cost-effective options for embedded systems is InfiniBand running on OpenVPX. The high-speed InfiniBand serial link – widely adopted in the larger commercial market (Sidebar 1) – allied with the high-bandwidth OpenVPX interconnection fabric is making waves as vendors tailor the technologies to the needs of the rugged, embedded world. Slowly but surely, as legacy bus-based architectures like VME run their course, a more powerful and affordable paradigm is ready to take their place.

Sidebar 1: InfiniBand, long adopted in the commercial market, is finding a niche in rugged computing.
(Click graphic to zoom by 3.0x)

VME versus VPX: Populating the ecosystem

The marriage of InfiniBand and OpenVPX in the embedded space was probably inevitable. The VPX point-to-point serial backplane interconnect is the migration path for widely deployed VME systems, shares form factors with VME cards, and boasts data rates at least 10 times faster than the newest incarnation of the bus technology. And, unlike VME, OpenVPX features a connector that supports high-speed serial links. InfiniBand can be configured in OpenVPX to provide a theoretically maximum throughput of 25 times the capacity of the fastest VME technology. That translates to a whopping 8 GBps data rate on the OpenVPX data plane.

OpenVPX/InfiniBand is an excellent option for image, audio, and signals intelligence applications. Moreover, it is ideally suited for radar because radar signals cannot be processed on one chip; they require multiple chips and therefore interprocessor communications. A bus-based radar subsystem would lack the throughput to keep up with more than a handful of CPUs. For anything beyond that, a switched fabric network topology – propelled by high-port-density switches – is necessary. An OpenVPX/InfiniBand solution potentially could perform more radar processing functions than a similarly sized VME system.

Vendors are speeding adoption by introducing InfiniBand-on-OpenVPX options at key niches in the embedded ecosystem, such as the switches that connect end point processors. These switches, which plug into OpenVPX backplanes, are the engines behind InfiniBand subsystems. They provide connectivity to InfiniBand-enabled processor cards, creating the VPX data plane network for interprocessor communications. One of the first rugged, 6U OpenVPX/InfiniBand switch fabric modules in the military/aerospace embedded market is GE Intelligent Platforms’ IBX400, with Double Data Rate (DDR) InfiniBand pumping out bits at 400 Gbps on the data plane and with 1 Gbps Ethernet on the control plane (Figure 1).

Figure 1: The GE Intelligent Platforms rugged IBX400 6U OpenVPX InfiniBand Switch Fabric Module is designed for High Performance Embedded Computing applications.
(Click graphic to zoom by 1.9x)

Compelling case

The combination of blazing speed, low latency, and scalability in the hardware with mainstream-tested, performance-boosting, zero-cost middleware makes InfiniBand-on-OpenVPX a compelling candidate for upgrades to VME-based systems as well as for new systems.