Engineers developing backplanes must cost-effectively achieve the appropriate level of performance for their application. When selecting a backplane technology, they must consider the performance options available, the realistic cost economies resulting from widespread deployment of the technology, and how much support and off-the-shelf software and hardware is available through the technology's vendor ecosystem.
Ethernet became the incumbent backplane technology for a wide range of storage, low-end wireless, wireline, industrial, and other embedded applications as engineers moved away from proprietary implementations. The availability of off-the-shelf hardware and software, as well as expert support, reduced development complexity and time-to-market. It also significantly lowered backplane development costs.
However, Ethernet was originally designed to connect a large number of client endpoints over a flexible and extensible architecture. Over the years, it has continued to work well in local-area network (LAN) and wide-area network (WAN) applications. Yet its inherent flexibility often makes it inefficient to use in high-speed embedded backplanes, especially as backplanes move from 1 Gbit/s to 10 Gbits/s, where the increased processing requirements exceed Ethernet's limits.
AN ALTERNATIVE TO ETHERNET The RapidIO standard has its origins as a next-generation front-side bus for high-speed embedded processors. Early in its development, its creators realized the value of an efficient front-side bus that could also serve as a system-level interconnect. Designed specifically for embedded in-the-box and chassis control plane applications, RapidIO provides minimal latency, limited software impact, protocol extensibility, and simplified switches achieving data rates from 667 Mbits/s up to 30 Gbits/s.
RapidIO technology isn't constrained by Ethernet's requirement to serve in a wide range of disparate applications. Ethernet's commitment to maintaining backwards compatibility also introduces inefficiencies since, for example, fields that aren't entirely necessary anymore like preamble and IFG increase header size. Additionally, Ethernet's generalized header means even simple services such as messaging or packet loss recovery mechanisms must be implemented as higher-layer protocols that increase header size and protocol processing complexity.
RapidIO achieves a far more efficient implementation since the base protocol supports many functions important in backplane applications, such as guaranteed delivery, read/write operations, messaging, data streaming, quality of service (QoS), data plane extensions, and protocol encapsulation, all without the overhead of higher-layer protocols.
With its more efficient header and integrated features, RapidIO also achieves better data bandwidth efficiency. Ethernet can improve its efficiency by sending larger packets, but this introduces undesirable latency jitter. Ethernet also can drop packets, requiring fabric over-provisioning. Assuming 25% to 35% usage, this limits Ethernet to a maximum effective throughput for layer 2 traffic of about 2.5 Gbits/s over a 10-Gbit/s link.
Ethernet relies on software for protocol processing to increase its flexibility. The processors required to process stacks at high data rates, though, significantly increase system cost and power consumption. Attempts are being made to improve Ethernet's bandwidth efficiency through hardware-based offload engines. Such engines, however, aren't implemented consistently across the industry. They require customized stacks that are difficult to maintain across product lines, increase system cost and complexity, and can affect product interoperability.
The RapidIO protocol is, per spec, implemented in hardware to guarantee optimal performance at the lowest cost. This additionally stabilizes implementations, guaranteeing consistency and interoperability across the industry, leading to more choices for developers.
Much of Ethernet's appeal stems from its high volumes and wide support. But most of this support is for LAN and WAN applications. Backplane Ethernet requires specialized functionality such as virtual LAN QoS and serializer/deserializer physical layers, significantly reducing the number of accommodating vendors, applicable market volumes, and therefore cost advantage. RapidIO enjoys wide industry support specifically for backplane applications. Furthermore, RapidIO silicon complexity and cost are comparable to Ethernet, yet RapidIO offers 2.5 times more effective bandwidth per link than Gigabit Ethernet.
RapidIO technology was designed to provide optimized performance in backplane applications. By offering higher bandwidth efficiency, integrated functionality, hardware-based protocol processing, and a strong ecosystem, RapidIO provides reliable interconnect technology that will carry engineers well into the future.
Reprints
