VMEbus: Forty and Fit!
Very few computer-related technologies exist in the industry for more than a few years. This past October 21st marked the 40th anniversary of the announcement of VMEbus. This event is especially significant given the fact that VMEbus is still an extremely practical solution for critical embedded computing. VMEbus has had a profound impact on the open standards community, playing a key role as an influencer in modular open system architecture (MOSA) initiatives.
Many standards for embedded computing owe their existence to the efforts of the pioneers of VMEbus: CompactPCI, AdvancedTCA, and VPX all exist due to VMEbus. Today’s innovators now have multiple avenues for developing industry standards. Technology has become so complicated that it is nearly impossible for any individual company to create and develop full product lines that meet the demanding needs of any single industry.
The early years of VMEbus saw the emergence of a push to open standards. The U.S. Department of Defense (DoD) was one of the early advocates of open standards. Key leaders recognized early the benefits of MOSA: that the only way to a robust and successful ecosystem of technology providers was through the use of open standards. MOSA requirements are now built into many program contracts.
Soon after the announcement of VMEbus, a VME Manufacturers Group was formed. In 1985, it became the VMEbus International Trade Association (VITA), chartered to accelerate the technical and commercial acceptance of VMEbus, and to help build a VMEbus supplier ecosystem. The first edition of a VMEbus product directory was released in May of 1983 with 196 listings from 45 companies (Figure 1).
The standardization effort for VMEbus struggled in the early years, primarily because a suitable development venue and process were not available. VITA eventually stepped up and added the VITA Standards Organization (VSO) to its charter, thus providing a home for the full stewardship of VMEbus and its future evolution. VITA evolved to become an industry-leading standards development organization dedicated to the creation of open standards for the critical and intelligent embedded computing industry. VMEbus has led to Futurebus, VPX, VNX, and a range of mezzanine standards.
VMEbus Systems magazine was launched in 1985 (Figure 2). The charter of the magazine was to cover the VMEbus industry; this scope was later extended to cover all of VITA standards and the related technologies under its current title VITA Technologies.
Influence on critical embedded computing
VMEbus found its greatest success in what is now defined as critical embedded computing. These systems must work flawlessly to protect life, property, equipment, and the environment. These are systems that must be “able” in many dimensions: dependable, supportable, configurable, reliable, serviceable, etc. The real-time deterministic capability of VMEbus made it an ideal solution for larger and more complex real-time computing systems, especially those needing a high degree of real-time control and data processing. The ability to scale both processing and I/O capability created a model that is emulated in today’s even higher-performance systems, like VPX. Backplane-based topologies like VMEbus enable architectures that are not possible in motherboard or “pizza box”-style configurations, ideal for complex and I/O intensive applications. VMEbus also pioneered open standards in rugged systems that can operate in extreme operating conditions. The experience gained in VME is being passed on to the next generation of backplane-based solutions.
VMEbus has been used in countless applications too numerous to list, but over the years, it has settled into a key component in many defense applications. Driven by its real-time performance capability and ruggedness, it is a favorite critical embedded solution to many platforms (Figure 3).
Setting the foundation
VMEbus was derived from the 68000-microprocessor bus, so it was a natural extension to use a 68K microprocessor on any processor board. The concept of single-board computers with processor, memory, and I/O all on the same board was still a few years away. A first-generation processor board had little more than the processor, boot ROM, and a serial port or two. This naturally led to the memory-board wars as companies raced to increase memory capacity (the highest-capacity boards only had 128 KB of RAM). Companies were competing to make the densest and most cost-effective memory cards possible.
By the late 1980s, the concept of a single-board computer was emerging. Memory density had reached the point where enough could be put on the same board as the processor that the board no longer needed to always have a separate memory card for the most basic of systems. In the area of I/O, Ethernet was just starting to appear, again requiring a separate and very specialized card just for the most basic connectivity, mostly to small dedicated systems as the internet was not connected to much at the time.
Disk storage was also a challenge. Most systems used simple storage systems that took a lot of effort to implement. Originally SASI was very popular, but the newly conceived SCSI bus was gaining attention. Still too complicated to implement without a dedicated VMEbus board, both battled for market position.
Mezzanine cards did not exist in the beginning. The first-generation mezzanine cards were proprietary to each vendor and primarily used to add more memory to a VMEbus slot. Mezzanines for I/O, such as Industry Packs and M-modules, started to gain the attention of the industry. The Sun Microsystems S-bus and IEEE PMC [PCI mezzanine card] fought it out for several years before PMC became the preferred choice.
Since those early days, most I/O functionality has been merged into the processor. Memory capacity and functionality are way beyond anything that could have been conceived in 1981. What used to take three to five VME slots moved to a single slot and now – in many cases – to a single chip, leaving most of the board space empty. Today’s VME boards have plenty of space on the PCB after going through decades of fighting for every square millimeter!
“When VMEbus was defined, the form factor seemed too small and early systems required multiple boards,” states Mac Rush, a retired board hardware design engineer at Motorola Computer Group. “But the entire electronics industry worked to miniaturize everything. Chips integrated more functions. Surface mount technology packages reduced the size of components. ASICs, PLDs and FPGAs allowed us to develop custom logic. Memory density increased. When I started, a 300 MB disk drive used three-phase power and was the size of a washing machine. Now a 256 GB drive is fingernail-sized. The board functionality I designed in the first days of VMEbus have been reduced to single chips with much more capability. VMEbus really benefited from the radical changes in functional density.”
VME going forward
The history of VMEbus has been widely documented: A full timeline and history can be found on the VITA website at . VMEbus is the only truly deterministic, real-time optimized, backplane-based solution in the industry. The scalable flexibility and rugged operation capability are second to none. While no one may be working on added improvements to VMEbus, knowing what we have experienced the past forty years, VMEbus can be ex-pected to live on for many more years.
VITA staff members’ musings on VME
All the VITA staff members have experienced careers heavily influenced by VMEbus. They were asked to comment on how VMEbus affected their careers.
Dean Holman, Assistant Executive Director, VITA
It was 1985 and I was a year out of college working as a systems engineer at MITRE Corporation. I was working on the Joint STARS program, attempting to verify if the proposed display processor design would have enough horsepower to meet the rendering needs of the downward looking tank tracking application. I was given a “vanilla” VME 19-inch rack, and some Motorola MVME-147 boards to use in the benchmarking study. At that time, the backplane was straight VMEbus across all slots. There were no interslot signal traces on the P2 connectors to facilitate interboard data transfers. If I wanted to use those user-defined pins on the P2 connector, I had to plug ribbon cables onto the back of the P2 connector shroud between the two slots of interest.
Then in the late 1990s and early 2000s, I was designing next-generation VMEbus boards and systems myself at Mercury Computer Systems. By then, the RACEway overlay on VME was in full swing. It allowed simultaneous switching of three streams of data traffic across the backplane, complimenting the VMEbus data transfers. These increases in bandwidth and clock speeds allowed for a quantum jump in system cross-sectional bandwidth and overall mission-processing capabilities.
Fast forward to today, and the interslot communications available to support VME is astounding. Mesh fabrics, star interconnects, full switching in a 21-slot chassis. There are very few limitations to the systems engineer with regards to how they can move data between the boards in a system today. VMEbus systems of today are still commanding a huge market share of systems, be those in the aerospace and defense, transportation (i.e., trains), or industrial automation segments. The longevity of VME after 30 years is impressive.
As data throughputs and processing needs continue to increase exponentially, newer system architectures such as OpenVPX, with their faster data movement capabilities and processing bandwidths, will continue to gain ground. I believe these newer systems will replace the legacy VME systems in some sizable percentage of the aforementioned segments in order to support the ever-increasing mission needs of the future. Those increased requirements are due to the vastly larger data sets made available via new sensors, information sharing between multiple platforms, and the need for artificial intelligence to be able to process Big Data. Autonomous vehicles – whether land-, sea-, or air-based – all require massive data movement and processing capabilities best met by the newer architectures. That said, the demonstrated ability for system designers to increase clock speeds and memory density will continue to allow the time-proven VMEbus systems in the field to be upgraded to meet numerous applications for many years to come.
Jing Kwok, VITA Technical Director
As young design engineer, VMEbus was a well-documented standard that made it easier for me to follow and hone my design skills. Having a well-documented standard supplied direction that would have been impossible to get otherwise. As a midlevel engineer, I had the opportunity to work with the standards community to develop new standards. I was able to take part in standards working-group discussions, eventually leading to a position as the editor for the early work on VPX.
As a senior engineer, I had opportunities to drive standards-based designs into corporations and the standards community. VMEbus gave me opportunity to meet with customers to educate them on the VME standards and ask for system feedback that drove internal design solutions which resulted in design wins. In my current role as the VITA technical director, I get to participate in a multitude of standards working groups, helping them to better define the standards and get them through the open standards consensus process under VITA.
Ray Alderman, Chairman of the Board, VITA
In the 70s, 8-bit microprocessors overwhelmed the markets using relay ladder logic and programmable controllers (sequencers) in low-end industrial applications. In the 80s, DEC, Data General, Prime, Micro Data, and some other minicomputer vendors were the 16-bit dinosaurs roaming high-end industrial applications, but again just as sequencers. That all ended when the 68000 processor, real-time operating systems, and VME came to market. By the mid to late 80s, upstart VME was the only the deterministic event-driven architecture and took the market away from the minicomputer dinosaurs.
To this day, VME is still the only deterministic real-time event-driven architecture in the market. Its proven capabilities in the industrial markets spilled over into telecom and the military markets in the 90s. Today, we have the serial fabrics that can move data around faster than VME, introducing data-driven architectures. But fabrics can never come close to determinism and real-time processing. So, VME put Ethernet on the P0 connector and added the MBLT transactions to move data around. VME can manage determinism on the parallel bus and move data on the fabrics, combining both event-driven and data-driven architectures.
The original designers of VME left “hooks” in the original specification, so we could hang new capabilities on them as technology advanced over time. That’s where we added 32-bit transactions, fabrics on P0, 64-bit (multiplexed) transactions, and MBLT. No other bus technology in history has done that without destroying backward compatibility. That’s why VME is still a viable and attractive embedded computer architecture today. No other computer architecture in history has stayed vibrant for 40 years. And VME will still be practical at 50 years. I see nothing on the horizon that can replace VME in real-time deterministic applications.