VME: No time to die, again

Story

November 19, 2020

By Robert Persons

The death of VMEbus – a technology used so heavily in military and aerospace applications – has, like the death of Agent 007, been predicted many times for many years, but all those premature death reports have never come to pass. Why?

As we eagerly await the final installment of Daniel Craig’s James Bond saga, I think back to late January 2020 when a group of us old-timers gathered for the annual Embedded Tech Trends conference, which is a gathering of COTS [commercial off-the-shelf] product manufacturers along with editors of well-known technical publications. I have attended this event for years and have seen, essentially, the same graying group of technologists and editors who come to discuss the future of COTS products we collectively sell to various industries, primarily for the military and aerospace sector. VME and COTS dominated those chats for decades and still do as they also still flavor each day’s presentations. During their talks, technology leaders show the assemblage the latest technologies, most of which are becoming smaller and more powerful to solve the future needs of the warfighter, whether it is a tactical system in a ship, an autonomous drone, or a CubeSat.

Inevitably we see a chart from a market research firm showing trends in the market for certain backplane technologies and we all anticipate the graph that compares VME versus OpenVPX/VPX sales projections. This is always the same time each year when we all look to see if the total revenue for VPX crosses the mystical total revenue for VME, which is nearly 40 years old. VPX still has not exceeded VME. (Figure 1.)

[Figure 1 | Graph compares sales of VME and VPX during the years 2014 to 2020. Source: IHS Markit via Jerry Gipper.]

Now I can hear everyone out there who questions the report, and I agree that total sales of all VPX/OpenVPX has probably surpassed VME by now. However, it is amazing that VME is still alive and kicking, and is still being used in a large number of programs, both in the military/aerospace and industrial sectors.

The story of VME

The story of VME started back in the 1970s a few years after the release of Dr. No, which saw Sean Connery (may he rest in peace) bring Bond to the big screen for the first time. Motorola began working on products based on an early bus called VERSAbus using a Eurocard mechanical standard. In 1981 (“For Your Eyes Only)”, Motorola coauthored the Rev A version of the VMEbus specification with Mostek and Signetics.

[Editor’s note: OpenSystems Media cofounder John Black coauthored the VMEbus specification while an engineer at Motorola.]

SMART Embedded Computing used to be Artesyn, which was previously part of Emerson, which bought Motorola Computer Group. The CTO at the time, Shlomo Pri-Tal, was one of the early architects of the specification, so VME is really part of the SMART DNA. The early bus was patterned after the Motorola Semiconductor 68000 processor. The goal was to design a specification that could be used to build embedded computer systems with a variety of I/O and multiple processors loosely coupled through a common bus. (Figure 2.)

[Figure 2 | From the archives: A photo of the most popular VME boards ever produced – the MVME147, a popular single board computer. Even though the board was launched 32 years ago, in 1988, SMART EC is still approached to this day at tradeshows and via the website by users wanting to buy it. Photo courtesy Jerry Gipper.]

If you look at the time this was happening, IBM PCs were just starting to populate the world and were really not ideal for industrial applications which needed a lot of I/O. The VMEbus standard was designed to allow for multiple processor boards and multiple I/O boards which was revolutionary at the time. Motorola, Heurikon, and Force, all companies that eventually became part of SMART Embedded Computing, along with a number of other companies, started building competitive products based on VMEbus.

The first death: Multibus II

Back during that same time, a competing standard, Multibus – developed by Intel and later adopted by Sun Micro­systems – was in early competition with VME especially with the release of Multibus II (Timothy Dalton’s first Bond outing in “The Living Daylights”). The cards were larger and could accommodate higher-powered processors and faster backplane speeds. Back in the day, we heard that Multibus II would kill off VME because of the superior performance, but the 6U Eurocard design became the de facto standard for embedded computing, and Multibus II eventually faded away. It is hard not to emphatically stress the importance of that early decision for the 6U Eurocard design. It was the perfect balance of board surface area and the ability to ruggedize. The size of a Multibus II board made that much more difficult to ruggedize and thereby limited its impact on embedded computing.

The original VMEbus specification was ratified by IEEE in 1987 after it was completed by VMEbus International Trade Association (VITA), which had formed in 1985 out of the VMEbus Manufacturers Group. This trade group was a pioneering group of early industrial-computing vendors who formed the organization to produce the standard and promote the industry to their collective customers. It was radical because all companies competed against each other, but they also needed one another to create and promote the standard. As we say today, even though it’s a cliché, co-opetition [or “cooperative competition”] created a stronger foundation for a system design than any single vendor trying to do it on their own.

The first release defined a bus protocol which enabled multiple masters or slaves on the bus with a blistering 40 MB/sec data transfer rate. VMEbus did not stand still: Subsequent enhancements to the standard started by first multiplexing address and data pins to increase the addressable space and the performance of the transfers. Direct-memory access (DMA) transactions were introduced, along with later enhancements to the bus clocking, pushing the transfer speeds to 80 MB/sec by 1994. By 1997, improvements to the standard increased the backplane performance to a respectable 320 MB/sec. All these modifications were done while maintaining interoperability to the original boards. Backwards compatibility was another hallmark of the VMEbus standard and is probably the main reason why it has survived so many competing standards.

The second death: CompactPCI

For VME’s second near-death experience, enter the CompactPCI (cPCI) standard in 1995, just as Pierce Brosnan appeared in “GoldenEye.” After Intel created PCI and it was adopted by IEEE, the concept of creating systems based on PCI became a focus. A new trade organization, the PCI Industrial Computing Manufacturers Group (PICMG), started for the very same reasons VITA did, but PICMG was for bused architectures based on PCI bus. Compact PCI is a 6U Eurocard-sized based board with a bus based on PCI. Along with the faster PCI bus, which was used for I/O, additional enhancements were included to make it appealing to the telecom industry. A time-division-multiplexed (TDM) bus was added, with eventually Ethernet in the backplane and integrated system management. With all these new features, it was thought that surely cPCI would knock out VMEbus. It was actually a bit of a joke between the standards groups that cPCI was going to kill VME. The irony was that many of the engineers that worked on VITA standards also worked on PICMG standards.

Although cPCI did make inroads into some of those traditional markets of VME, especially in the smaller 3U versions of the product, several things kept VMEbus strong. One was momentum: VMEbus was now more than 10 years old. An ecosystem had formed, major programs were using VME, and it was designed to be a multiprocessing bus from the outset. PCI bus really followed a master-slave paradigm, one leader controlling a group of followers. There were multiprocessing cPCI systems, but this required special nontransparent PCI bridges, and in most cases, this was not something that could be done dynamically. You had to have a special version of a processor board that was a slave board, or the board had to be jumpered as a master or slave. Multiple masters was a key design element of VME and it was that capability that was heavily used by the programs that based their systems on VMEbus. This setup gave older programs a way to add features to an existing computer system by adding a new processor board to perform expanded functions.

The third death: servers

As cPCI came on the scene, Intel-based servers also began to expand. Surely this could be a potential place where VME could finally meet its demise, specifically in the industrial sector, especially since a typical system was comprised of a single master and some number of I/O modules. But VME dodged the bullet.

For sure, there has been a trend away from VMEbus in certain industrial applications, but surprisingly there still exist a number of customers who still buy a lot of VME. But why?

There are several reasons. Many of these customers have designed and built their own specialized I/O VME cards. They do not want to have to redesign these cards and are willing to use a bus that is more than 35 years old to maintain these designs. They have built up a semirugged system that in many cases is very close to or is actually physically attached to the pieces of equipment; VME is ruggedized and can live nicely in these extreme environments.

Actually, the industrial applications that are still using VME are those that continue to rely on multiprocessing in their applications. They have also designed the system to accommodate a wide range of applications with few I/O to ones that require a lot. They can adjust the performance/cost of the system by selecting different processor cards with different levels of performance and cost. Users who need more processing capability than the fastest processor card can add an additional one to offload. This does not mean that there hasn’t been a decline (in general) of VME in industrial applications – there actually has been – but there are still some key customers who rely on the flexibility of VME to this day.

The final nail in the coffin?

Then came VITA 46 and VPX.

The need for high-speed serial connections between processors, GPUs, FPGAs, and storage drove the VITA community to create a new set of standards in 2007 (in between the “Casino Royale” remake and “Quantum of Solace”). Later, as the plethora of slot and module profiles proliferated, there came a push to increase interoperability by developing a new standard, VITA 65, OpenVPX resulted. (Figure 3.)

[Figure 3 | Sales of VME and VPX – broken out by boards and systems – during the period 2014-2020. Source: IHS Markit via Jerry Gipper.]

Certainly, this would indeed be the final blow to VME. Well, maybe it eventually will be, but there are a number of factors where VME is still able to avoid the knockout punch. When will the curve showing net sales of VPX products increasing finally cross the VMEbus curve as it declines?

OpenVPX/VPX products used in new designs – especially with the efforts being promoted by organizations like the Sensor Open Systems Architecture (SOSA) consortium, which promotes standardizing around limited OpenVPX profiles – are quickly driving the OpenVPX/VPX sales curves, but VME will still have some life into the future.

There are many military programs and industrial control programs that remain reliant on VME because it is good enough and any change would have major impacts to their systems. One factor that has driven the continued use of VME is the difficulty of replacing backplanes and cabling in existing designs – it is very costly to remove the chassis and recable. This is so often true for older existing designs, but it also impacts those newer systems that also rely on cabling and backplanes designed years ago. It is actually quite surprising how many programs are affected by this one fact: There are very few customers willing to pay to redesign their entire system. Enhancements of capabilities must be done through improved VMEbus processors and/or I/O cards.

The government typically doesn’t budget to update just the computer system in some of these programs. They may pay for incremental upgrades to processing to add a feature or budget for a complete replacement. Many of the VME sales today are people either trying to keep a system going by replacing a component that has gone EOL [end-of-life], replacing a processor card in a new build for a processor that has gone EOL, or adding features to an existing design and so buy a higher-performance processor.

Component obsolescence is only part of the challenge

A particularly tough challenge is the age of the application and underlying operating system (OS). Many years of software work has been done based on older operating systems, which then drives the integrator managing the design to purchase older VMEbus-based boards that are still in production because that older board supports an EOL operating system. Difficult decisions have to be made when replacing an existing board in light of the impact to the OS required and the lack of support for that old OS.

If the integrator has been updating their software development environment over the years, they are more likely able to use a newer board design with better longevity. If not, they are forced to stay with an older design, if it is even still available. One challenge – which came up around the same time as James Bond battled Blofeld in “Spectre” in 2015 – was the EOL for a widely used VMEbus chip. Many VMEbus board vendors had to decide at that point how to support their existing customer base. Some left the VMEbus market altogether, while others decided to either redesign the products with older VMEbus bridges or create an FPGA-based bridge. (Note: SMART Embedded Computing made a major investment to maintain its existing designs for as long as possible.)

So, like the character of Bond (James Bond), VMEbus will live forever, right?

No. As new programs replace older ones and organizations like SOSA start to push the industry into a more refined choice of OpenVPX board profiles, the momentum of these newer technologies will accelerate even faster. It will be a struggle for some who have relied on VME for so long, but even great technologies have to eventually become fond memories. Inevitable conclusion: As the old-timers retire in dribs and drabs, the industry will see the retirement of this important bus standard too. As Sean Connery was to 007, VME was the grandfather of all open standards. It may seem like an antiquated technology by today’s standards, but it was truly a technology superstar in its day and will always be an important part of embedded computing history. The real question now is: Who will be the next person to play James Bond? 

Robert Persons is senior sales architect for SMART Embedded Computing. He applies his extensive knowledge of embedded real-time systems, VMEbus and ATCA hardware, and real-time software to help SMART Embedded Computing customers accelerate their projects. His 30-plus-year career has included avionics software development and field support of military, aerospace, telecom, rail, and industrial customers. Rob has also represented companies on standards bodies and conference advisory boards; he holds dual bachelor of science degrees in computer science and zoology from the University of Central Florida.

SMART Embedded Computing
https://www.smartembedded.com