BROWSE BY TECHNOLOGY



RTC SUPPLEMENTS


SOLUTIONS ENGINEERING

VME 25th Anniversary

VME Market Bifurcation: The VXS and VPX Face-Off

With the advent of two new specifications, the VME market might appear to be bifurcating. While true in a sense, the reality is that two complementary technologies are becoming available to offer developers even more choices in performance, flexibility and ruggedness.

ERAN STROD, MERCURY COMPUTER SYSTEMS

  • Page 1 of 1
    Bookmark and Share

VME has enjoyed robust growth since it was adopted in 1981. Today the merchant VME board market is north of $1B and growing at a respectable clip. In aerospace and defense, nearly 80% of the merchant boards procured are VME. Furthermore, VME is expanding in this market at nearly 8% CAGR for the foreseeable future. There is a perception that VME is in retreat in commercial markets, but this is not entirely the case. Electronic Trends Publications forecasts growth in VME in the industrial computing market at more than 7% CAGR through 2010.

Interestingly, despite much merger and acquisition activity, the top players in the VME merchant market have remained relatively stable. Table 1 shows the top VME board vendors. Mercury was named the top VME board vendor for 2005. VME has enjoyed continual renewal as newer and better technologies are added to it in an evolutionary fashion. Over the past 25 years, the data transfer rate over the tried-and-true DIN connector has increased by an order of magnitude; the latest innovation being VME320. VME64x added a fabric connector that further increased inter-slot data rates. VME’s success has partly depended on having maintained backward compatibility at the pin level.

VXS (VITA 41) vs. VPX (VITA 46)

The most recent tension in the VME market relates to VXS (VITA 41) and VPX (VITA 46). Some customers are concerned the market is bifurcating. They are right! The market is bifurcating, but not in the way that is commonly thought. VXS and VPX actually have a great deal in common.

Figure 1 shows the typical architecture of a sensor computer used in a radar or signals intelligence application. The system has a radio frequency module to transmit and receive signals, an A/D board, a quad-processor card, and a single-board computer with dual-mezzanine I/O slots. The boards are interconnected with a high-speed fabric that is depicted using a fabric switch card.

In the software domain, there is system software, data movement middleware, numeric libraries such as VSIPL and of course, the user application. Note that this canonical system is form-factor-agnostic. It could be VXS. It could be VPX. It could be something else completely. All of these form-factors use the same basic technological ingredients: PowerPC, RapidIO, Linux, etc. In fact, Mercury has demonstrated families of products based on VXS, VPX-REDI, AdvancedTCA, CompactPCI and other form-factors—all based around a common processor/fabric/software architecture. With a single underlying software and hardware architecture, the ultimate form-factor deployed can reflect an application designer’s needs and preferences in mechanical robustness and ecosystem.

VXS – VPX Differences and Similarities

The mechanical differences in VXS and VPX are fairly straightforward if you read the standards. The differences are summarized in Table 2. VXS is exactly like VME64, maintaining the P0 and P2 connectors, but adds an improved P0 connector that supports modern multi-GHz serial fabrics like RapidIO. With a VXS backplane, system engineers can carry forward VME64 cards without the necessity of a hybrid backplane. Note, the VXS P0 obsoletes the VME64x P0. This is a difficult development for VME64x customers who are now coping with the fact that the leading VME64x fabric silicon company appears to no longer be a growing concern.

Unlike VXS, VPX is a ground up redesign of the backplane chassis. VPX can support VME, but does not mandate that VME be carried forward. It eliminates the VME P1 and P2 DIN connector in favor of the new multi-GHz connectors. As shown in Table 2, it has much greater I/O pin capacity. When teamed up with the REDI standard (VITA 48), the result is truly revolutionary, supporting higher slot power budgets, enhanced ruggedization and allowing for an array of open-standard cooling methodologies: liquid flow through and liquid spray in addition to conduction and air.

It’s natural to think that VXS and VPX will compete with each other if you focus only on the mechanical differences between them. In reality, they complement each other. This can be understood only if we examine the ecosystems of the respective standards. Each standard addresses a different set of application requirements—one “demanding” and the other “extreme.”

Over time, system designers will always gravitate toward the simplest solutions possible. For example, if a single Pentium motherboard running Linux will solve a problem at hand, why look further? If the application is fairly straightforward, designers will typically select a benign solution (Figure 2). Benign applications are generally stationary, run in temperature-controlled environments, require one or only a few processors and modest connectivity. These solutions play an important function in the overall ecosystem by setting a baseline for price and performance. Further, they improve relentlessly, forcing competing solutions to follow a similarly aggressive continuous improvement curve.

Rugged VME Systems

As embedded designers know, commodity technologies are great, but not if they can’t get the job done. Some applications require NEBS compliance (telecommunications) or must be capable of surviving shock, vibration or temperature extremes (defense). Some applications have to scale to a large number of I/O ports or require numerous inter-connected processors. When an application can’t get by with “benign” product, system engineers usually adopt more rugged technologies (Figure 2). VME’s bread-and-butter niche, where VXS is assuming the leadership mantle, has been in 6U rugged applications. Air-cooled CompactPCI and AdvancedTCA are more robust than the benign alternatives, and that’s fine for the telecom central office, but for mobile applications like defense and transportation, VME still offers a dominating value proposition.

In rugged computing, customers tend to adopt technology at the board level. With hundreds of VME module vendors, customers have an enormous ecosystem from which to choose. The VITA Web site lists 24 categories of VME modules. These offer a wide array of processors, field programmable gate arrays (FPGAs), interfaces and functionalities. With an open-standard backplane, customers can choose best-in-class modules from numerous vendors and then integrate them together. This provides customers with enormous flexibility and choice.

The vendor competition in the VME space means that new technologies are made available very quickly. For example, processor MHz increases are constantly being fielded on new boards as vendors jockey for performance leadership. The evolution of technology around an open-standard backplane is rapid.

Example VXS System

A very high-performance VXS system available today, Mercury’s PowerStream 6100 system, can house 68 PowerPC processors for 761 GFLOPS and supports more than 42 Gbytes/s of fabric bandwidth. That represents an order of magnitude performance leap over what we could do just two short years ago. Is it enough? For many applications it is, but there are some customers who can never have enough processor performance, fabric bandwidth, I/O connectivity or ruggedization. These customers adopt technologies in the extreme computing category (Figure 2). Extreme computing is a category in which the application requirements exceed what can be supported by simply integrating off-the-shelf boards.

A large prime contractor came to Mercury with a challenge: “We need 125 PowerPCs in a ruggedized air-cooled chassis the size of a medium-sized refrigerator.” Later, a customer asked us to fit 69 Xilinx FPGAs in this same box. These challenges could not at the time, and still cannot today, be accomplished with VME. In order to solve the problem, system architects had to start with a clean sheet of paper. The backplane fabric was based on parallel RapidIO. The system was so densely packed with processors that Mercury invented a technique to cool it called “managed air.” Mercury developed its own backplane architecture and form-factor called MultiPort. The result is a system-level product called PowerStream 7000, which is being used on the Multi-Platform Radar Technology Insertion Program (MP-RTIP), among other major programs.

The PowerStream 7000 is an example of a system solution where processing, fabric, I/O and software are all designed holistically. You can’t take out one module and use it somewhere else—everything in the system is designed interdependently. Detractors of this approach will characterize it as proprietary, but that is not a realistic view. The system is completely based upon open standards: RapidIO, PowerPC, PCI Mezzanine Cards, POSIX, etc. Customers who require extreme levels of performance, whether that involves processing, I/O connectivity or ruggedization, simply have had to look beyond the open backplane PICMG and VITA systems. Today that situation is changing with the advent of VPX (VITA 46) and REDI (VITA 48).

VPX-REDI

The extreme computing application segment will soon be able to adopt VPX, an open-standard backplane architecture, which provides processing and connectivity for the industry’s most demanding applications. VPX supports 768W per slot. In first-generation systems, power is actually more limited by cooling capability than by inlet capacity. Thus, VPX is particularly revolutionary when paired with the Ruggedized Enhanced Design Implementation—REDI (VITA 48). With REDI, designers have open-standard mechanical specifications for four cooling methodologies: air, conduction, liquid spray and liquid flow through. REDI supports many other ruggedization enhancements that enable systems to address the harshest environments on earth or beyond.

In order to address the diverse needs of extreme applications, the team working on VITA 46 has defined a range of options. Table 3 describes the major options available to designers of VPX-compliant solutions. A large number of options is good, of course, because it preserves flexibility. However in practice, there are so many options for VPX that it makes one wonder about the likelihood of third-party vendor interoperability. There are a few vendors who have announced VPX product to date. It is too early to know, but it is possible the small community of VPX vendors will select different form-factors, fabrics or mechanical specifications such that modules cannot be mated together in a system. Still, large program adoption is driving VPX solutions toward Q4 2006 availability.

As is natural in the early phase of technology adoption, the pioneering VPX-REDI vendors are making an investment to establish the market. In practice, this means most VPX-REDI vendors are building system-level product. Why? Because there is not yet an ecosystem from which to draw. For example, Mercury is integrating A/D, processing, FPGA and I/O functionality together in a conduction-cooled VPX-REDI architecture. This VPX-REDI system is called the PowerStream 6600. Since there is no VPX-REDI ecosystem to tap, Mercury is building the entire 6600 system with enabling software from the ground up. Based on experience with the extreme computing market, Mercury is confident the VPX-REDI ecosystem will emerge over time.

Electronic Trends Publications forecasts that VPX products will start to ship in 2006 and grow to exceed $30M by 2010. It sees VXS growing to exceed $80M in that same time frame. Both standards will thrive, but VXS is widely expected to have a running start because it more easily leverages the VME-board ecosystem. Someday, VPX-REDI will certainly have matured enough so customers will be able to adopt at the board-level and plug modules from different vendors into the same system. After all, that’s what an open-standard backplane is all about. The industry will get there, but a brand new technology like VPX-REDI requires a bit of time for the ecosystem to crystallize.

What if an application requires more than VPX-REDI can offer? That’s the market that the MultiPort form-factor addresses. Now that VPX-REDI is coming to market, some customers will demand an order of magnitude better performance than even VPX-REDI can deliver. Future systems will deliver an enormous performance leap using the Cell Broadband Engine—a binary compatible G5 PowerPC processor with vector engines from IBM. There are today and always will be applications for which there is never enough performance and connectivity no matter how much COTS vendors deliver. The goal of the MultiPort family is to innovate a generation ahead of the PICMG and VITA standards, even as the standards organizations continually improve their offerings.

The real bifurcation in the VME market is between a la carte VXS board customers who want to perform their own module integration and VPX-REDI system customers who adopt technology at the holistic system level. This bifurcation will affect many different aspects of the VXS and VPX-REDI markets. In VXS, since adoption is at the modular level, pricing models will be unbundled. In the VPX-REDI market, software, hardware and services will be sold as a bundle. As a board market, the VXS ecosystem will enable customers to choose best-in-class modules from different vendors and integrate them together. Anyone who has spent time in the lab knows that integration of complex systems involving multiple vendors takes significant resources. Adopting VPX-REDI technology at the system level trades vendor flexibility for quicker time-to-market. VXS customers may be comforted in knowing that large vendors are now offering module-integration services to assist customers in defraying the significant effort of inter-vendor module integration.

There is a temptation to regard VXS and VPX-REDI as competing with each other, but in reality, they have different capabilities; and in the first generation of products, will draw upon their respective ecosystems differently. VXS is emerging as a modular-board standard that addresses rugged applications that cannot be satisfied with benign computing solutions. VPX-REDI will emerge as a system-level standard, with an ecosystem of a select few, highly committed vendors. VPX-REDI will go beyond rugged computing and address the needs of applications with extreme performance and ruggedization requirements. Both form-factor architectures will stand as distinct and viable alternatives, even as they share fabrics, processors, I/O interconnects and software.

Mercury Computer Systems
Chelmsford, MA.
(978) 256-1300.
[www.mc.com].