Futurebus looked to fix these problems and create a successor to systems like VMEbus with a system that could grow in speed without affecting existing devices. In order to do this the primary technology of Futurebus was built using asynchronous links, allowing the devices plugged into it to talk at whatever speed they could. Another problem that needed to be addressed was the ability to have several cards in the system as "masters", allowing Futurebus to build multiprocessor machines. This required some form of "distributed arbitration" to allow the various cards gain access to the bus from any point, as opposed to VME which put a single master in slot 0 with overall control. In order to have a clear performance benefit, Futurebus was designed to have the performance needed ten years in the future.
Typical IEEE standards start with a company building a device, and then submitting it to the IEEE for the standardization effort. In the case of Futurebus this was reversed, the whole system was being designed as during the standardization effort. This proved to be its downfall. As companies came to see Futurebus as the system, they all joined in. Soon the standards meetings had hundreds of people attending, all of them demanding that their particular needs and wants be included. As the complexity grew, the standards process slowed. In the end it took eight long years before the specification was finally agreed on in 1987. Tektronix did make a few workstations based on Futurebus.
That was just in time for the US Navy who had been looking for a new high-speed system for the Next Generation Computer Resources (NGCR) project for passing sonar data around in their newly designed Seawolf class submarines, and they said they would standardize on Futurebus if only a few more changes would be made. Seeing a potential massive government buy, the additions effort started immediately on Futurebus+. This also had the unexpected side effect of killing any effort to produce Futurebus system while everyone waited for the new version to come out, "real soon now". Real soon turned out to be another four years, and when the resulting Futurebus+ was released, no one was interested any longer.
All of the Futurebus+ proponents had their idea of what Futurebus+ should be. This degenerated into "profiles", different versions of Futurebus+ targeted towards a particular market. Boards that were compliant with one Futurebus+ profile were not guaranteed to work with boards built to a different profile. The Futurebus+ standards development politics got so complicated that the IEEE 896 committee split from the IEEE Microcomputer Standards Committee and formed the IEEE Bus Architecture Standards Committee (BASC).
In the end very little use of Futurebus was attempted. The decade-long performance gap they gave the system had evaporated in the decade-long standards process, and conventional local bus systems like PCI were close in performance terms. Meanwhile the VME ecosystem had evolved to such a degree that it continues to be used today, another decade on. The Futurebus technology is currently used as an internal backplane technology for systems such as routers.
However the Futurebus effort did act as a catalyst for change in other ways. After the 1987 version shipped and the Futurebus+ effort started, a number of the original designers realized the effort was doomed. One member did a quick back-of-the-envelope calculation and showed that by the time Futurebus+ shipped, it would already be too slow for the supercomputer market. A group then organized to create a system aimed directly at this need, which eventually led to Scalable Coherent Interconnect. Meanwhile another member decided to simple re-create the entire concept on a much simpler basis, which resulted in QuickRing. Due to the simplicity of these standards, both standards were completed before Futurebus+.
Futurebus was the source of some of the original work on Cache Coherency, Live Insertion of boards, and Trapezoidal Transceivers. Trapezoidal Transceivers have a controlled risetime and make backplane and bus design much simpler. The original Trapezoidal Transceivers were made by National Semiconductor. Newer Futurebus+ transceivers that meet the IEEE Std 1194.1-1991 Backplane Transceiver Logic (BTL) standard are still made by Texas Instruments. Futurebus+ was used as the I/O bus in the DEC 4000 AXP and DEC 10000 AXP systems. Futurebus+ FDDI boards are still supported in the OpenVMS operating system.
Futurebus systems were implemented with 9Ux280 Eurocard mechanics using 96-pin DIN connectors resulting in a backplane that supported both 16 and 32 bit bus widths.
To understand Futurebus+ you need to read many IEEE standards;
896.2 contains three Profiles for target markets, A for general purpose systems, B for an I/O bus, and F for a Futurebus+ will all the options that will make it go fast. Profile A was sponsored by the VMEbus community. Profile B was sponsored by Digital Equipment and implemented in VAX systems as an I/O bus. Profile F was sponsored by John Theus while he worked at Tektronix and was intended for high end workstations.
Futurebus+ supports bus widths from 32 to 256 bits. It is possible to build a board that supports all of these bus widths and will interoperate with boards that only support a subset. Split bus transactions are supported so that slow response to a read or write will not tie up the backplane bus. Cache Coherence, implemented using the MESI protocols, was very complicated but significantly improved performance. Futurebus+ was one of the first open standards to support Live Insertion which allowed boards to be replaced while the system was running.
Futurebus+ boards are 12SUx12SU Hard Metric size defined in the IEEE 1301 standards.
One of the most elegant features of the Futurebus design is its distributed bus arbitration mechanism. See US patent number 5060139 for more information. In the end this was replaced by a central arbiter.