Traditionally, the telecom central-office computing market has been served by numerous providers with proprietary solutions. This has a few implications: System development is expensive, the products are inoperable with other vendors' products, time-to-market takes longer, and upgrades can be costly and prohibitive.
But the PCI Computer Manufacturers Group (PICMG) has introduced the Advanced Telecom Computing Architecture (ATCA) to alleviate these problems and offer exciting new levels of performance in an open standard.
ATCA, implemented over the PICMG 3.0 core specification, targets markets where CompactPCI (cPCI) encounters performance problems. First, the architecture is geared to meet the challenges of a telco environment. The 8U by 280-mm form factor and 1.2-in. slot spacing provides much more real estate on the server blades for components. The ZD connector it uses handles speeds of up to 5 Gbits/s. Using various switched fabrics for the traffic engine, performance can hit gigabit/terabit levels, with high availability and scalability.
There are four sub-specifications: PICMG 3.1 (Ethernet), PICMG 3.2 (InfiniBand), PICMG 3.3 (StarFabric), and PICMG 3.4 (PCI Express). The PICMG 3.5 subcommittee recently formed to develop RapidIO-based fabric and node boards for ATCA. System management, direct dual-redundant 48-V dc power distribution, and cooling capacity of 200 W per board are important requirements of the specification.
But will ATCA be supported? Well, over 100 companies helped develop the specification, and Intel has been a strong backer, supporting the move to the all sheet-metal chassis. The company has supplied a lot of the muscle on defining the specification, as well as its marketing and industry prowess. Lucent and Motorola have also been active players. At recent ATCA Interoperability Workshops, over a dozen companies demonstrated the interfunctionality of cards, backplanes, chassis, and so forth.
On the hardware side, backplane and chassis vendors like Bustronic, Elma Electronic, Pentair, and Carlo Gavazzi have already developed 12U platforms, mostly with 14-slot Dual Star (two hub switches running the fabric) or Mesh (each slot acts as a hub slot) backplanes. Some, like Elma, have also announced 4U horizontal versions with five-slot Mesh backplanes. Development units in 2U height with three-slot backplanes should hit the market soon.
Another important consideration is that little else in an open-standard architecture will run PCI Express and Express Advanced Switching adequately. PCI Express is Intel's and the PCI Special Interest Group's major initiative to replace PCI.
ATCA is widely expected to succeed, and a few companies are betting the farm on it. While Gigabit Ethernet and StarFabric can be used across the cPCI architecture, PCI Express needs higher performance. Because cPCI has problems with speeds over 1.3 Gbits/s, ATCA is one of the best ways to handle PCI Express' 2.5-Gbit/s performance at the moment.
Of course, proprietary solutions can often handle the higher speeds, but as noted, they have disadvantages. First is the high cost of prototyping and low- to medium-volume production. Even for high-volume programs, ATCA offers advantages. As an open standard, it has dozens of vendors for competitive pricing, and a single-source, proprietary solution entails more risk. Compatibility is another issue. With a whole community serving the ATCA technology, most hardware is interoperable and the mechanicals, software, pin assignments, etc., are all defined specifications. This is not true when trying to mix and match with proprietary systems.
Geared for telecom central offices, AdvancedTCA is expected to be successful. With the new form factor, direct 48-V dc power, built-in redundancy and reliability, and massive performance potential, ATCA may revolutionize the market. However, the best technology does not always win. Timing is always an important factor. We'll have to wait and see.