The new battlefront: open architecture wars

Predicting which of these embedded computer standards will succeed is not possible, but the features and positioning of the AdvancedTCA, VXS, and Embedded Modular architectures can give us some hints. One can expect that VME will remain solid and strong for many years and its current offerings will suffice for a wealth of applications. For the applications needing more performance, VXS should provide a strong transition. In the telecom arena, Embedded Modular could give AdvancedTCA competition for high-end systems, depending on the features and level of cooling that systems demand.

The VMEbus has been going strong for over 20 years, and has continuously expanded performance from 16-bit J1 VME to 64-bit VME64 extensions, to the VME320. However, some applications such as systems that are sensor- and DSP-card intensive often need higher performance or more bandwidth than VME provides. Several military/aerospace, medical and even some industrial applications fit this criterion.

For high-performance VME, a couple of new architectures are compelling. The first is VXS, which is fully compatible with VME64x. VXS adds a high-speed connector over P0 of a VME64x backplane for serial data traffic. Designers will have the flexibility of plugging in standard VME64x cards for parallel-bus only, integrating new payload and switch cards for parallel bus and switch-fabric transport, or switch-fabric transport only. The VXS spec allows for four differential serial pairs per direction link over P0, and supports up to two such ports on each VMEbus card.

VXS should be attractive for most of the applications where VME is a good fit, but more performance is needed. As VXS is fully compatible with VME, the introduction to switched interconnects such as Infiniband (VITA 41.1) and Serial RapidIO (VITA 41.2) is not as drastic a change. Users will get all of the benefits of VME, plus the high-speed dataplane using the fabric of their choice. Other fabric implementations in the near future may include StarFabric, Gigabit Ethernet, and PCI Express.

Embedded Modular Architecture (VITA 34) on the other hand, is VITA's stab at offering an architecture that can cross over more into the often highly regarded communications space. The architecture is based on 4U x 220-mm or 8U x 220-mm cards or may use a more standard 6U x 160-mm size option.

With fears of lightning-quick processors building up tremendous heat, an aggressive liquid cooling approach is being implemented. Each individual card may be enclosed in a metal case to accommodate this type of cooling, as well as provide intense shielding. Although well suited for high-end defense and industrial applications, the specification may reach a wider audience.

With tremendous bandwidth, excellent cooling capability, and stringent EMI/RFI control, VITA 34 may become a solution that crosses over to communications applications. The specification will use a high-speed connector for the data traffic, with promises of up to 10 Gbit/second performance, with much higher aggregate bandwidth.

VITA 34's progress has been very slow, but it may turn out to be well timed. By the time it is ready, the communications market may be strong and we may have a lot more need for the aggressive cooling and shielding performance. However, we'll have to see if the cost and feature-base suit a large market or just a smaller niche.

In the PICMG arena, CompactPCI (CPCI) has been a key backplane-based champion. It grew rapidly from its introduction in 1998 until the telecom crash. With higher pin counts, Eurocard form factor and hot-swap capability, it has been a great open specification for communications systems.

With 33 MHz processors limited to eight slots over CPCI backplanes (without bridging) and 66 MHz limited to five slots, the PICMG 2.0 specifications started to run into its performance limitations. The PICMG 2.16 and PICMG 2.17 specifications, running Ethernet and StarFabric fabrics respectively, have boosted performance and reliability and will extend CompactPCI's life. But the PICMG organization realized its market needed even higher bandwidth and saw the telecom central office as a key segment. Therefore, AdvancedTCA was introduced in late 2002.

AdvancedTCA is the major initiative from PICMG, with over 125 members participating. The 8U x 280-mm cards and 1.2-inch pitch allow large server blades with a wealth of components to be used. Using primarily dual-star and mesh switched-fabric topologies, the architecture will be able to handle interfaces up to 40 Gbits/s (for Terabit backplane bandwidth), high availability (99.999 percent uptime), and quality-of-service issues demanded by the telecom central office. The backplane allows for 48 Vdc input from an external source to be distributed to the individual slot cards.

PICMG sponsors hope that base AdvancedTCA will become an all-encompassing architecture for network architectures from the data center to the core to the access edge. Sub-specifications under Advanced TCA for Ethernet (PICMG 3.1), Infiniband (PICMG 3.2), StarFabric (PICMG 3.3), and PCI Express (PICMG 3.4) have been ratified. A new subspecification for RapidIO over AdvancedTCA has been announced as PICMG 3.5.

AdvancedTCA looks strong for next-generation telecom. However, cooling the 150 watts to 200 W per board and incorporating shelf management are going to be problematic. Also, the specification calls for 48 Vdc for power, so it will be a hassle for development and demonstrations.

Melissa Heckman is an electrical engineer at Bustronic Corp. (Fremont, Calif.).

See related chart