GE Fanuc Embedded Systems

VPX: VMEbus for the st 21 Century

VPX: VMEbus for the 21st Century

The VMEbus architecture has served military embedded computing extraordinarily well since the VMEbus specification was first published in 1981. VMEbus brought industry standards to a computing world then dominated by proprietary architectures and operating systems, and was in the vanguard of the movement that saw the much more open computing environment of today. This paper looks at VPX, the latest VME standard and how it brings new levels of computing performance to the embedded military applications of today and tomorrow. What VME brought to embedded computing What strengths did VMEbus bring? It was an open standard, with no single company having control of the specification. Its architecture was specifically optimized for real-time computing, and it offered a full 32-bit data path and 32-bit addressing. It could support multi-master CPU configurations to boost performance and processing bandwidth. It has had a high bus bandwidth. Backplane I/O gave it significant configuration flexibility and high maintainability. Rather than using edge connectors, it used pin and socket connectors for greater robustness and reliability. It adopted a standard Eurocard board size: 3U (160 x 100 mm) or 6U (160 x 233.35 mm), a format used throughout the industry thus making e.g. the material needed for cage assembly easy and cheap to obtain. VME and VME64 became industry standards (IEEE-1014 and IEC821). VMEbus brought modularity and upgradeability where previously systems were typically monolithic and inflexible. Its focus on rigorous standards and its high performance made it popular, and a virtuous circle saw an enormous ecostructure of vendors grow up, with hundreds of products and ready availability of support, further feeding its popularity. But while a key design goal of the VMEbus standard was to anticipate the future in as many ways as possible, it was always unlikely given the nature and speed of change in the computing industry that it would last forever. Those who worked on the original specification have subsequently expressed surprise that it lasted for ten years let alone the 25+ years it has been at the heart of embedded computing. Technology and applications change A development that was impossible to foresee, for example, was the advent of serial switched fabrics such as Serial RapidIO, Infiniband, PCI Express and Gigabit Ethernet, and how these would revolutionize the performance potential of embedded computing. Rapidly increasing integration of functionality on individual devices, while more foreseeable than serial switched fabrics, created levels of power consumption and thus heat dissipation that went beyond those envisaged by the original VMEbus specification. At the same time, military computing requirements were becoming significantly more demanding with the focus on network-centric warfare, increasing use of graphics and
2

video, digital signal processing, sensor data acquisition and processing, and generally more compute-intensive applications. Network-centric warfare may be a term that has become over-used, but it accurately describes a future in which combat forces will rely on, and become part of, a network that constantly delivers superior situational awareness. At the heart of this network will be distributed, network-centric computers that rely on the power and performance of multiple processors in order to receive, interpret and distribute a broad range of sensor-derived information. There are identifiable challenges as military computing moves forward (Figure 1). Radar processing, for example, requires massive bandwidth. Edge interfacing is a requirement in sophisticated Software Defined Radio systems. The image processing associated with surveillance gives rise to a requirement for advanced inter-nodal performance. New generations of peripherals feature higher speed interfaces. Challenge Massive bandwidth Edge interfacing Inter-nodal performance High speed peripherals I/O signalling frequencies Small form factor Typical Application Area Radar Processing Software Defined Radio Surveillance image processing DVI, Serial ATA, 10G Ethernet FPGAs Unmanned Vehicles

Figure 1: New military application challengers

The increasing popularity of FPGA technology places enormous demands on I/O signaling frequencies. These levels of performance cannot be achieved within the limitations of todays VMEbus-based systems with their restricted I/O and bus bandwidths. Enabling VMEbus-based systems to leverage the capability of switched fabrics is fundamental to allowing the armed forces to achieve the required computing power. Bringing VME up to date Thus, a new standard was required that responded to these significant changes in application requirements and technology. One of the fundamental strengths of VMEbus has been its ability to adapt to these changes over time, yet maintain mechanical and electrical compatibility with earlier implementations. VME64, for example, provided support for a 64-bit bus plus a third backplane connector with 95 contacts for rear I/O in 6U-sized cards. An early attempt to respond to the growing popularity and availability of serial switched fabrics was made by VITA the VME International Trade Association, a body that has been largely responsible for ensuring the continuing relevance and competitiveness the VMEbus architecture with the VITA 41 standard, which became known as VXS. The VXS definition

VPX: VMEbus for the 21st Century

replaced the 2mm hardened connector introduced with the VME64 definitions with a new high-speed connector capable of hosting Gigabit Ethernet, PCI Express, and Serial Rapid I/O serial links in the backplane. The problem facing VITA was, however, that to really achieve significant bandwidth and performance improvements, and to optimally accommodate the specific requirements of serial switched fabrics, a new connector and board form factor definition was required but a new connector and board form factor would, by definition, be physically (if not electrically) incompatible with the existing VMEbus infrastructure. The connector challenge Traditional connector technology worked well with the parallel data buses that have been widely used notably in VMEbus systems. However, newer switched fabrics use high-speed differential signalling (2.5 Gbits/second or higher per serial link), and give rise to a need for a different kind of connector. Historically, connectors have incorporated pins arranged such that the vertical and horizontal pitch positions of the contacts are symmetric. Its an approach that works well for single-sided signals. However, the differential signals that characterize switched fabrics operate better over pairs of pins that are close to each other, but that are as distant as possible from other pairs. Add to this the increased requirement for good grounding that arises as a result of much faster signaling speeds, and the need for a new connector technology becomes apparent. A compromise had to be struck, and a poll indicated that lack of physical compatibility with legacy systems was an acceptable price to pay for the required level of performance, given that all the other advantages and benefits of VMEbus (open standard, same fixed card sizes, electrical compatibility) would be retained. However, to accommodate the needs of those wishing to be able to upgrade existing VMEbus-based systems with VPX

Figure 2: The hybrid chassis allows VMEbus users to integrate VPX boards

technology, or for those wishing to construct hybrid systems that leveraged the capabilities of both architectures, the concept of a hybrid chassis (Figure 2) was mooted and exposed to major organizations within the defense industry. Reaction to the proposal was positive, and the principle of a hybrid chassis, capable of supporting both VMEbus and VPX boards, is now accepted as being a key element in allowing existing VMEbus users to integrate the capabilities of VPX into existing systems. VPX: the beginning A working group was convened by leading defense and aerospace suppliers (Figure 3) in mid-2002 to begin discussion of the proposed VITA 46 standard, the primary goal of which would be to preserve VMEbus users investment in hardware and software while delivering a migration path into

Figure 3: The VITA 46 working group comprised many of the industrys leading players
3

VPX: VMEbus for the 21st Century

the future. The working group comprised, among others, Boeing, Curtiss Wright Controls Embedded Computing, GE Fanuc Embedded Systems, Mercury Computer and Northrop Grumman. One of the key reasons for the growth in adoption of VMEbus-based systems was the inherent robustness and reliability of the connector so it was inevitable that this part of the VPX specification would come under close scrutiny. It was submitted for independent (by Contech Research Inc.) and exhaustive reliability qualification testing in order to establish its suitability for harsh military environments. Tests performed by Contech included shock; random vibration; bench handling; humidity; salt fog and sulphur dioxide; dust and sand; durability with misalignment; ESD; and current overload. The final test report was submitted to VITA by Contech in October 2005: the ruggedness of the Tyco MultiGig RT2 was found to be exemplary, with similar performance under stress conditions to that of the existing VMEbus connector. Power and cooling The other key improvement that VPX brings to the VME world is in the area of provision for the increased power consumption of todays highly integrated silicon. While each VMEbus slot was limited to a maximum of 90 watts at 5 volts, VPX allows for up to 115 watts at the same voltage, or up to 384 watts at 12 volts or 768 watts at 48 volts. This increased power budget allows VPX board designers significantly more flexibility in selecting and placing devices. However, the issue of cooling still has to be addressed more so, given the increased power consumption possible with VPX. The specification of 6U VPX calls for a conduction-cooled envelope compliant with IEEE-1101.2, which is compatible with existing enclosures. Provision is also made for air-cooling via an IEEE 1101.1/10 form factor version.

Figure 4: The new Tyco MultiGig RT2 connector

The new connector At the heart of the VPX standard as VITA 46 became known is a new, high bandwidth connector. The MultiGig RT2 from Tyco (Figure 4) is a 7-row, 16-wafer (wafers can be power, differential or single-ended) that delivers highly controlled impedance, minimal insertion loss and less than 3% crosstalk at transfer rates up to 6.25 Gbits/second. A 6U VPX board features six 16-column 7-row RT2 connectors and one 8-column 7-row RT2 connector, while a 3U board features two16-column 7-row RT2 connectors and one 8-column 7-row RT2 connector. The new connector enables a VPX board (Figure 5) to feature a total of 707 non-power electrical contacts and a total of 464 signal contacts. The latter are usable as: 64 signals implemented as 32 high speed differential pairs for core fabric 104 VME64 signals 268 for user I/O including 128 high speed differential pairs (giving a total of 160 high speed differential pairs) 28 for system utilities or spares The connector was chosen to allow a typical stiffening bar and a standard length PMC, ensuring that current generations of PMC and XMC and legacy PMCs and XMCs remain compatible with the VPX environment.

Figure 5: VPX board section showing new connector, alignment/ keying block

VPX: VMEbus for the 21st Century

REDI For more stringent cooling requirements, the working group came up with the REDI (Ruggedized Enhanced Design Implementation previously known as VITA 48) standard which describes how to implement layout techniques to better support cooling methodologies on specific form factors. REDI is a physical, rather than electrical, standard and while it has often been closely associated with VPX it arose from the same working group it is independent of VPX and can be implemented for alternative architectures. REDI provides a specification for the design of forced air, conduction- and liquid-cooled implementations, and also addresses spray cooling. To allow for greater power and heat dissipation, REDI includes provision for increased board-to-board spacing and increased board thickness. REDI also makes important provisions that, together with the improved ESD (electrostatic discharge) characteristics of the new RT2 connector, allow for two level maintenance. Two level maintenance becomes possible Historically, maintaining electronic equipment in the field has presented the armed forces with significant and expensive challenges. Three- or four level maintenance paradigms have given rise to complex and inefficient working practices that have been identified as requiring rationalization. Already in place in some programs and environments, the militarys goal is to move to a universal two level maintenance (TLM) paradigm which would see much greater responsibility and capability for repair placed in the field, rather than at a depot. For this to happen, however, requires the deployment of systems which can readily be handled in environments that are less than ideal there must, for example, be no risk of damage by ESD (electro static discharge), sand, dust and the like. In helping the military achieve this ambition, the Tyco MultiGig RT2 connector has a role to play: it features inbuilt ESD protection via a grounding strip. Beyond this, REDI specifies board covers (Figure 6) that provide additional protection from environmental factors. The combination of the two allows field replacement of faulty modules within a chassis (rather than replacement of the entire chassis) with minimal risk, making two level maintenance a practicable approach. Some estimates put the savings that accrue from moving to two level maintenance as high as 40%.

Secondary side cover

Primary side cover

Board

Figure 6: ESD covers fitted to either side of a VPX board

3U VPX enables new applications Another important benefit that VPX brings to military application developers is in the high performance achievable in its 3U implementation. For several years now, developers have been frustrated by the performance limitations of 3U VME and this has led to VME losing ground in perhaps the fastest-growing part of the market for smaller, lighter systems, and the growing implementation of 3U CompactPCI. Perhaps the most important disadvantage of the 3U implementation of VME was its lack of rear I/O pins (none for 3-row VME connectors, and only a minimal number for 5-row connectors), thus requiring I/O to be placed on the front panel where space is often limited: 3U CompactPCI, on the other hand, can provide up to 75 pins of I/O per system slot, and up to 105 pins per peripheral slot. Beyond this: where 3U VME64 could achieve a maximum peak bandwidth, for example, of 40 Mbytes/second, 3U CompactPCI could achieve up to 528 MBytes/second. The data bus width of 3U VME is up to 16 bits: CompactPCI allows for up to 64. However, CompactPCI is not without its limitations, especially in environments where high frequency data is being transmitted that has a high requirement for data integrity. Many observers believe that VPX, with its superior pin count and provision for high reliability and high speed data transfer, will offer a solution that will lead military systems designers to prefer it over the CompactPCI 3U alternative.

VPX: VMEbus for the 21st Century

Figure 7: GE Fanuc 3U VPX SBC330 Intel Core Duo-based Single Board Computer

GE Fanuc Embedded Systems and VPX GE Fanuc Embedded Systems was one of the companies that participated in the VITA 46 working group whose discussions led to the formulation of the VPX standard. As such, its not surprising that the company was among the first to bring VPX products to market or that the company now has perhaps the broadest range of VPX solutions available, with no fewer than nine products announced and more in the pipeline (Figure 7). VPX single board computers The 6U SBC610 is based on the Freescale 8641D dual core PowerPC processor. With up to 2 GBytes of memory, it offers unrivalled I/O flexibility with its support for both on-board and plug-in I/O via its two XMC/PMC sites and unique AFIX site. It enables a range of scalable solutions from single board systems to large multiprocessor systems, offering both high speed fabric connectivity (sRIO and PCI Express) and legacy connectivity (VME). The SBC610 is fully supported by Radstones market-leading AXIS Advanced Multiprocessor Integrated Software tools and libraries. Also based on the 8641D PowerPC, and with the option of either single- or dual core, the 3U SBC330 benefits from the improved signal integrity and power management offered by the VPX form factor. Each PowerPC core, operating at up to 1.5 GHz, has 1 MByte L2 cache and an AltiVec vector
6

processor. The 8641D can support symmetric or asymmetric multiprocessing (depending on operating system support), and its dual 64-bit DDR2 memory controllers can be assigned to each core for increased operating system isolation, or can be shared between the cores. The 8641D natively supports two 8-lane PCI Express ports. One of these is routed directly to the backplane, where it may be used to connect to peripheral devices, such as the GRA110 graphics processing unit. The second PCI Express interface routes into a PCI Express switch, from which four ports are routed to the backplane and the remaining lanes provide interconnect to USB and Serial ATA interfaces. The SBC340 3U VPX single board computer is based on the Intel Core Duo processor. Implementing the 2.0 GHz T2500 processor architecture, the SBC340 brings state of the art processing performance to the military and aerospace market. With a rich set of I/O, the SBC340 is a rugged single board computer aimed at processing, communications and display applications. The Intel 945GM Northbridge provides a dual-bank DDR SDRAM interface, as well as incorporating GMA 950 graphics capability. High-speed interfaces include Gigabit Ethernet, high speed USB 2.0, SATA and GPIO. The VPX form factor allows for high speed PCI Express connections to other cards in the system, with 16-lane and 4-lane PCI Express ports supported across the backplane.

VPX: VMEbus for the 21st Century

VPX multiprocessing Part of GE Fanuc Embedded Systems 6U VPX product family, and designed with support for the AXIS Advanced Multiprocessor Integrated Software in mind, the DSP230 offers scalability from four to eighty PowerPC AltiVec nodes per enclosure. High speed data movement between I/O and processor nodes is simple to implement using the AXISFlow Inter Processor Communication (IPC) module. AXISFlow leverages the DSP230 message passing accelerator architecture to ensure non-blocking inter node and inter board data movement. The DSP230 exploits a unique dual fabric architecture with both Serial RapidIO and PCI Express between all nodes and system I/O as well as Gigabit Ethernet connectivity to the network. VPX graphics With the inclusion of NVIDIAs leading edge G73 Graphics Processing Unit (GPU), the 3U GRA110 graphics board brings desktop performance to the rugged market. With outstanding functionality, together with PCI Express interconnect, the VPX form factor allows for high speed connections to single board computers in the system. The GRA110 supports the 16-lane PCI Express implementation, providing the maximum available communication bandwidth to a CPU such as the GE Fanuc Embedded Systems SBC340. The PCI Express link will automatically adapt to the active number of lanes available, and so will work with single board computers in 8- and 4-lane configurations. With a rich set of I/O, the GRA110 is designed to serve many of the most common video applications. Dual, independent channels mean that it is capable of driving RGB analog component video, digital DVI 1.0, and RS170, NTSC or PAL standards. In addition, the GRA110s video input capability allows integration of sensor data using RS170, NTSC or PAL video formats. VPX storage The SDD910 is a high-performance, high-reliability, solid state disk drive module providing large capacity storage for rugged 3U VPX systems running applications such as simulation, digital mapping or terrain visualization. Based on state-of-the-art NAND Flash technology, the SDD910 offers storage capacities up to 128 GBytes. The module offers burst read and write speeds up to 150 MBytes/second, and features sustained read and write speeds of 44 MBytes/second and 40 MBytes/second. The SDD910 implements error detection and correction algorithms to ensure data integrity. To protect sensitive mission data, the modules are capable of sanitizing the entire contents of the disk in seconds. The sanitization can be initiated under software control. For additional security, the SDD910 automatically resumes an initiated sanitization following a power interrupt. Two SATA interfaces are provided at the VPX backplane: the primary interface is used for the on-board solid state disk, while the secondary interface is routed to a SATA connector mounted on the front panel, allowing an external disk to be connected to the system. VPX switches Allowing designers to build complex VPX systems with multiple single board computers and multiple I/O modules, the PEX430 provides both a PCI Express switch and a mezzanine carrier. The PCI Express capability of the PEX430 allows up to four ports of 4-lane PCI Express to be connected to a nonblocking line-speed switch. The PEX430 switch supports non-transparent bridging mode to allow multiple hosts to be connected together. The PCI Express switch also has an 8-lane PCI Express port to the XMC site, or alternatively a 4-lane PCI Express port to a PCI-X bridge for the PMC site. The NETernity GBX410 is an IPv6 capable fully managed Layer 2/3 Gigabit Ethernet switch in 3U VPX form factor designed to meet the most demanding requirements for network switching in tactical applications. Featuring a nonblocking shared memory architecture that provides a 44 Gbits/second core and delivering full wire speed performance with minimal latency to all ports simultaneously, the switch is available in both air- and conduction-cooled formats. The GBX410 has comprehensive management capabilities that include VLANs, Link Aggregation, Spanning Tree, IPv4 and IPv6, traffic policing, QoS, guaranteed bandwidth and SNMP support. This switch can be expanded to a 24-port non-blocking solution by connecting two GBX410 together via the integral 10G uplink port. VPX subsystems Bringing unprecedented graphics performance to the rugged marketplace is the GE Fanuc Embedded Systems MAGIC1 Rugged Display Computer. Combining the processing power of the Intel Core2 Duo with the NVIDIA G73 GPU, connected together with 16-lane PCI Express, the MAGIC1 Rugged Display Computer is capable of driving the industrys most demanding visual applications. The processing node consists of a Core2 Duo CPU running at 2.0 GHz with 2 MBytes of L2 cache. System memory is made up of two banks of dual data rate DDR2 SDRAM, with capacity up to 4 GBytes. The graphics processing node is based on the dual channel NVIDIA G73 graphics processing unit, as featured on the NVIDIA GeForce 7600GT, incorporating 256 MBytes of GDDR3 SDRAM arranged in two banks. The interface between GPU and CPU is 16-lane PCI Express, allowing maximum bandwidth between the two processors. Storage is provided by a solid state disk drive, which boasts a

VPX: VMEbus for the 21st Century

capacity of up to 64 GBytes, sustained read performance of 45 MBytes/second, and a purge facility to allow data on the drive to be securely deleted in an emergency. The MAGIC1 Rugged Display Computer is packaged in a base-plate cooled chassis capable of operating in the harsh environments that characterize many military and aerospace applications. Summary The VMEbus architecture has served military embedded computing applications well for over a quarter of a century. Increasingly demanding applications, coupled with new technologies such as serial switched fabrics, meant that a new derivative VITA 46, now VPX was required that leveraged many of the familiar characteristics of the VMEbus architecture but brought new levels of performance. While introducing some elements of incompatibility with what has gone before, VPX brings substantial improvements in price/performance, especially for applications where size and weight are constrained. It seems likely that VPX can serve the needs of military embedded computing for the next quarter of a century.

GE Fanuc Embedded Systems Information Centers Americas: 1 800 322 3616 or 1 256 880 0444 Asia Pacific: 86 10 6561 1561 Europe, Middle East and Africa: +49 821 5034-0
2007 GE Fanuc. All Rights Reserved. All other brands or names are property of their respective holders.

Additional Resources For more information, please visit the GE Fanuc Embedded Systems web site at:

www.gefanucembedded.com

06.07 GFT-645