Electronic Design

Advanced TCA: A High-End Switch Fabric That Costs Less

Design by committee resulted in a range of standards with a common backplane and support for Gigabit Ethernet, InfiniBand, StarFabric, and PCI Express.

Creating the multiple specifications for this newly approved standard was a massive undertaking with contributions from over 100 companies. It brought designers a great deal of choice, especially when it comes to the backplane interconnect. Advanced TCA targets the carrier-grade telecom market. But thanks to its large board size and flexible switch-fabric architecture, it also is ideal for military and IT environments. That said, products for the new Advanced TCA (Telecom Computing Architecture), or ATCA, standard aren't a guaranteed market, especially with the slow return of the telecom market. However, the open nature of the architecture makes it a desirable platform for companies looking to lower the cost of previously proprietary, high-performance products.

Unlike the 3U and 6U 160-mm CompactPCI, ATCA addresses higher-end systems with its larger 8U 280-mm Front Board format that handles up to four Common Mezzanine Cards or PCI Mezzanine Cards. Of course, there's already discussion of a smaller CompactTCA standard. This would be more of a direct competitor with the existing PICMG 2.16 Ethernet and 2.17 StarFabric CompactPCI switch-fabric specifications.

The PICMG 3.0 base standard addresses the physical portion of ATCA. It goes into great detail about the hardware, including cards, connectors, backplane, subrack, shelf, and shelf management. Additionally, the base interface features Gigabit Ethernet connections. These are always supported in a dual-star configuration with redundant hubs. The Shelf Management Controllers are connected to this network.

The other specifications address the switch fabrics that can be used. These include Ethernet, InfiniBand, StarFabric, and PCI Express Advanced Switching. The base standard defines the type of board interconnects that each fabric must use. This includes a full-mesh, dual-star, and a dual-dual-star interconnect. The base specification will handle up to a 16-slot rack: Two slots are reserved for hub cards, and the remaining 14 are node cards. In a full-mesh interconnect, all 14 nodes have a direct connection to their siblings. A dual-star is the minimum configuration. In this case, one star is dedicated to the switch fabric and the other to the base Gigabit Ethernet interface. The dual-dual-star configuration has four hub slots.

Three zones make up the connectors on the back of each card (Fig. 1). Zone 1 supplies a -48-V power connection and the shelf-management network interface. Zone 2 provides the data transport support for the switch fabric, while the optional zone 3 is for user-defined connections. Normally, it contains connectors for the rear transition modules (RTM) so that it can handle cabling to devices on the main boards. This approach is similar to other architectures like CompactPCI.

A major feature of the ATCA base specification is the shelf-management and Intelligent Platform Management (IPM) support. Each board and most major devices in a case, such as the fans and power-supply modules, will have an IPM Controller (IPMC). A dual, redundant inter-integrated circuit (I2C) bus connects the IPMCs (Fig. 2).

The Shelf Managers configure the system when it starts. They recognize and configure added boards while the system is running so that the boards properly use the switch fabric and other interfaces available and track the health of the system using sensors installed within the system. The latter tends to be vendor-specific, but the application programming interfaces for the Shelf Manager are part of the standard. Thus, network management applications can access this information.

The Shelf Manager also participates in system power management. It obtains the requirements of each device in the system and controls the power consumption of on-board devices. Usually this support is incorporated into the operating systems that run on the entire system. All hardware on an ATCA board doesn't start running when the board is plugged into a hot system, or a system is powered on. Instead, each portion of hardware is controlled by its IPMC and, in turn, the Shelf Manager. So, power can be enabled to all or just sections of a board.

Shelf management can also be performed using Internet Protocol capable transport systems like Ethernet. This allows ATCA boards to support protocols such as Simple Network Management Protocol, Remote Management Control Protocol, and Web-Based Enterprise Management, along with network services like remote boot, iSCSI, and other high-level protocols.

ATCA boards are big and won't be hot when populated with the latest and fastest hardware. It's critical that a system's thermal design is up to the task. The standard allows the use of convection cooling, but most systems will employ fans or blowers.

Maximum power dissipation per slot is 200 W. A board that spans more than one slot can support 200 W per slot occupied. The RTMs are limited to 5 W per slot.

Creating the ATCA specifications was well worth the effort. It remains to be seen whether or not sufficient need exists to keep all related companies busy, and if all of the switch fabrics will survive in this space. In the future, ATCA may even support additional switch fabrics. For now, the standards appear to provide the best to all involved—a standard platform with a range of competing suppliers. Check out PICMG's Web site at www.picmg.org for details.

3.0 Defines overall mechanics, board dimensions, power distribution, power and data connectors, and system management plus 10/100/1000BaseT Ethernet connections
3.1 Ethernet fabric (dual-star or mesh)
3.2 InfiniBand (dual-star or mesh)
3.3 StarFabric (dual-star or mesh)
3.4 PCI Express Advanced Switching (dual-star or mesh)

Hide comments


  • Allowed HTML tags: <em> <strong> <blockquote> <br> <p>

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.