The mixture and flexibility of different fabrics and functions make MicroTCA interesting for a number of different applications. Also appealing is the ease with which one can migrate from other standards to MicroTCA without changing drivers or applications.

In addition, new products are coming onto the market that enable PCI and CompactPCI applications to benefit from MicroTCAís advantages. In most backplanes GbE over a switch and Serial Advanced Technology Attachment (SATA) as a slot-to slot connection are implemented. All other Low Voltage Differential Signaling (LVDS) lines in the backplane can be used for any of the available fat pipes, that is, PCIe, Serial RapidIO, and 10 GbE. The fabric in use in the backplane depends only on the plugged-in AdvancedMC modules and the switch on the MicroTCA Carrier Hub (MCH). Therefore the same backplane can be used for different fabrics and different applications without any modification. The question becomes, ìWhat application demands which fabric?î

For applications in dissimilar systems, Ethernet is often used, as it simplifies the communication between these applications. In some of these GbE transports control information and application data. If more bandwidth is needed, GbE is used as a low-performance control plane, and in addition 10 GbE is used as a data plane. For host-centric applications with existent PCI drivers, PCIe is recommended. As PCIe is backward compatible to PCI, the drivers can be reused without any modification. Therefore the same application runs on MicroTCA, CompactPCI, and PCI systems.

For multiprocessing applications with high bandwidth and low latency demands, Serial RapidIO is the first choice. This choice makes particular sense because TI and Freescale PowerPC DSPs implement Serial RapidIO on-chip, as do Altera, Lattice, and Xilinx FPGAs. As data packets and maintenance packets can be sent by any CPU to any other CPU or I/O or to the Serial RapidIO switch itself, implementation is much easier than with PCI Express.

The performance and flexibility of all these fabrics depend on the plugged-in AdvancedMC modules, but also on the used switch chip of the MCH. This will be discussed in more details for each of these fabrics.

Performance factors

The performance of a GbE link in a MicroTCA system depends on the GbE switch used on the MCH. One solution could be implementation with an FPGA or with a low-cost Ethernet Switch chip. Both of these solutions soon reveal their limits, if one needs to prioritize the data packets and/or needs the high bandwidth and low overhead on the CPU. If the MCH uses a manageable, nonblocking GbE switch, such as the Broadcom BCM5396, a lot of functions and buffers are available to allow prioritization, nonblocking data transmission, higher bandwidth, and lower CPU load. The key functions of such a switch are:

n  Nonblocking 17-port fully integrated GbE switch fabric

n  Sixteen SGMII/SerDes interfaces

n  Support of 9 KB jumbo frames

n  256 KB packet buffer memory

n  VLAN 4k 802.1Q or port-based support

n  Support of up to 4k unicast MAC addresses

n  Support of MAC-based port aggregation (trunking)

n  Flow control

n  Full-duplex (802.3x) and half-duplex options support

n  Support of automatic address learning and aging

n  Support for Spanning Tree and Rapid Spanning Tree

n  Support of 802.1x MAC Security

n  Port-based rate control feature with 64 Kbps granularity

n  Classifies packets using four 802.1p Quality of Service (QoS) or DiffServ priority queues

802.1Q tag Virtual Local Area Network (VLAN), trunking of ports, and 802.1p priority enables the switch to be designed into a wide variety of applications from unmanaged to websmart to managed switch. The Media Access Controllers (MAC) on the BCM5396 also support jumbo frames, which are typically used for high-performance connections to servers because they offer a smaller percentage of overhead on the link for more efficiency. The Broadcom Ethernet switch on the NAT-MCH is also suitable for fully managed applications due to the support for Spanning Tree and IEEE 802.1x MAC security. It is easy to configure with the Webserver-Interface of the NAT-MCH (see Figure 1).

Figure 1: N.A.T. MicroTCA visualization and control Java tool, NATVIEW (click graphic to zoom by 1.9x)

XAUI

For IP applications, which use the TCP/IP or UDP protocol, GbE is the best choice as no time is wasted for protocol transformation. If more data throughput is needed, 10 GbE is the logical upgrade step. The 10 GbE chip on the MicroTCA Carrier Hub aids system performance. The NAT-MCH employs for the 10 GbE switch function the Fulcrum MF2220, with 20 ports of 10 GbE L2 switches. 12 ports of this switch are connected to the 12 AdvancedMC slots in a MicroTCA system. One port is used as an update channel to the second MCH. The NAT-MCH Generation 3 also supports also two 10 GbE uplinks at the front panel as an option.

Figure 2: The NAT-MCH-base12-XAUI supports 10 Gigabit Ethernet to 12 AMC slots and two uplinks at the front panel. (click graphic to zoom by 1.3x)

Serial RapidIO

In applications with farms of processors, DSPs, and/or FPGAs, Serial RapidIO can be seen as a high-performance alternative to 10 GbE because of the guaranteed latency time and the hardwareís built-in error detection. The bandwidth of Serial RapidIO is also 10 Gbps per link. The aggregate bandwidth of the Serial RapidIO switch TSI578 is 80 Gbps.

n  RapidIO technology offers these advantages over GbE and 10 GbE:

n  Low CPU overhead

n  Higher effective bandwidth

n  Superior QoS

n  Superior latency and latency-jitter characteristics

n  Competitive or lower cost

As noted earlier, the on-chip implementation of Serial RapidIO on the TI DSP chips, Freescale PowerPC, and a number of FPGAs has increased Serial RapidIO popularity. In a MicroTCA system, from each AdvancedMC slot up to four Serial RapidIO lanes (10 gigabit) can be routed to a Serial RapidIO switch chip on the MCH. In a redundant MicroTCA system, four additional Serial RapidIO lanes can be routed from each AdvancedMC slot to the second MCH. Therefore in a system with two NAT-MCHs, up to eight SRIO lanes per slot can be used for data traffic. To support all 12 AMC slots with 4 SRIO lanes two Tundra TSI578 SRIO switch chips are built into the NAT-MCH. These are full mesh, nonblocking switching fabric chips with lookup tables for every port.

The NAT-MCH Generation 3 also supports two SRIO uplinks at the front panel as an option. And the TSI578 supports multicast. In computing-intensive applications the input data can be transferred simultaneously to all or a group of CPUs/DSPs/FPGA cards in a MicroTCA system.

PCI Express

Board drivers as well as hardware must be considered. How does one move an application and I/O boards from a PCI or CompactPCI platform to a MicroTCA system? In the current PCI and CompactPCI systems drivers are available for the I/O boards. To save time these drivers do not have to be rewritten to interface to Serial RapidIO or GbE. In these cases the PCIe fabric in the MicroTCA system is the right selection. PCIe technology is a low cost, highly scalable, switched, point-to-point, serial I/O interconnect that maintains complete software compatibility with PCI. The transfer rate is 2.5 Gbps per lane and per direction, and adding lanes to the link enables proportionate scaling for performance. The PCI Express protocol features are:

n  Boot existing OS without change

n  Scalable performance

n  Advanced features including QoS, PCIe hot plug support, and data integrity

The NAT-MCH-base6-x24 (Figure ?) has one PCIe Switch PEX8532 to support four PCIe lanes to each of six AdvancedMC slots (total 24 lanes). For the support of four PCIe lanes to each of 12 AdvancedMC slots the NAT-MCH-base12-x48 uses two PLX PCIe switches. This also allows clustering. The switches provide nonblocking switching with full line rate. QoS is provided, supporting two virtual channels and eight traffic classes per port. One of each port can be configured as a transparent upstream port, and one of each port can be configured as a nontransparent upstream port.

Now we have discussed all fat pipe fabrics and which one should be selected for which kind of application. Beside the higher bandwidth of the fabrics and compared to older standards MicroTCA offers the simultaneous usage of different fabrics in one system on one backplane.

Figure 3:  Telecom application with MicroTCA (MicroTCA Solver): 4 NAMC-16ADSP, 2 NAMC-STM4, 2 NAMC-8560-8E1/T1/J1, NAMC-8560-IO (click graphic to zoom by 1.9x)

Summary

Therefore MicroTCA allows the possibility of integrating with instead of replacing older systems and is a good solution for extending the life of older systems. So many applications can be realized with MicroTCA, that the name Micro Computing Architecture would better indicate this standardís versatility. MicroTCA offers the following.

n  More bandwidth (Serdes at 1-12+ Gbps)

n  Simultaneous data transfers without conflicts

n  Multicast and broadcast

n  Flexibility and adaptability of the infrastructure without software changes

n  Reduced costs for remote firmware upgrades and maintenance

n  Lower costs for inventory management of hardware in the field

n  Smooth and step-by-step upgrade of older technology

n  Long life cycle

n  Robustness

n  300 mm nominal equipment depth

n  Cooling of up to 40W (single-width AdvancedMCs) or 80 W (double-width AdvancedMCs)

With the new product family XLINK from N.A.T., PCI and CompactPCI-based applications can also benefit from the advantages of MicroTCA.

Dipl. Ing. Vollrath Dirksen is Strategic Business Development Manager at N.A.T. Before N.A.T. he worked as Key Account Manager and Senior Channel Manager at Motorola Embedded Communications Computing and Blue Wave Systems in Loughborough/UK in the telecom, industrial, defense, and medical market. Vollrath started his career as service engineer for Computer Tomography and then worked as technical trainer for microprocessors, digital signal processors, and real-time operating systems at Freescale and HILF! GmbH.

N.A.T. GmbH

www.nateurope.com
vollrath@nateurope.com