By John McHale
SAN DIEGO - Engineers at Arizona Digital in Scottsdale, Ariz., and General Micro ystems in Rancho Cucamonga, Calif., demonstrated VMEbus technology that delivers more than ten times the performance of existing VMEbus systems during the Bus & Boards conference last month.
The demonstration featured burst data transfer rates in excess of 640 megabytes per second using a standard VME320 backplane and standard VME64/PCI bridge components.
"Reports of the demise of VMEbus have been greatly exaggerated," says Ben Sharfi, president of General Micro Systems. "CompactPCI may be an up-and-comer at the low end, but it was designed primarily as an I/O bus, and will never be a match for VMEbus at the high end. Not only is VMEbus four times faster than CompactPCI, it delivers performance over 21 slots (vs. eight for Compact PCI), supports true multiprocessing, and provides must richer interrupt and bus control."
"It comes down to simple backplane physics," says Drew Berding, president and chief executive officer of Arizona Digital. "Nonreflective buses like VME320 will always be faster than reflective buses like CompactPCI, which act as low-impedance transmission lines and are inherently limited by ringing, noise, and reflection delays. With our VME320 backplane, which substitutes at star topology for the traditional daisy chain VMEbus wiring scheme, we eliminate those effects, making it possible to support sustained data transfer rates of up to one gigabyte per second while providing full interoperability with legacy VMEbus cards."
At the Bus & Boards conference next year Berding will have VME running at 1 gigabyte per second, predicts Ray Alderman, executive director of the VME International Trade Association (VITA) in Scottsdale, Ariz.
The question is will designers want those speeds if they do not need them? Berding, the inventor of VME320, believes they will.
The VMEbus system in the demo consisted of a 21-slot VME320 backplane designed by Arizona Digital, a General Micro Systems 255 single-board computer, a frame buffer card, a video card, and miscellaneous old VMEbus cards.
During the demo, data transferred between the frame buffer and video cards a complete line at a time at a burst rate faster than 528 megabytes per second. This data displayed on the high-resolution video monitor until each frame of the video was drawn. In addition, the fast transfer left lots of time for the CPU board and the frame buffer and other legacy VME cards.
In 1997 Arizona Digital released VME320, a modified VMEbus architecture. The VME320 architecture was designed for operation at more than 320 megabytes per second, and peak bandwidths faster than 500 megabytes per second. VME320 uses a new backplane design and bus protocol.
The VME320 backplane uses a star interconnection method to speed the VMEbus backplane itself. All of the interconnections on the backplane connect at the middle slot of the backplane.
In a nutshell, the leading edges of signals effectively pass through only one slot on the way to their destination by propagating from slot-to-slot, VITA officials say. This enables tight skew delays on the backplane, thereby speeding up the system.
The VME320 concept uses a new bus protocol called 2eSST. The protocol allows VME320 compatible modules to interact at much higher speeds.
Bustronic Corp. in Fremont, Calif., was the original licensee of the VME320 backplane. General Micro Systems officials expect to have silicon ready for the VME running at 533 megabytes per second later this year, Sharfi says.
To take advantage of the gigabyte per second data transfer rates made possible by VME320 backplanes, General Micro Systems engineers are working on a new single-chip VME64/PCI bridge chip that will provide 10 to 15 times the performance of other boards, they claim.
Leveraging 2eSST technology from Arizona Digital, the chip will support sustained VMEbus rates of 528 megabytes per second, an artificial limit imposed by the 64-bit 66-MHz PCI side of the bridge. Later, as PCI local bus speeds increase, General Micro Systems engineers will ramp up the chip's VMEbus performance, offering standard VMEbus transfer rates in excess of one gigabyte per second, Sharfi says.
"Cost has been a major stumbling block for VMEbus," Sharfi says, "but most of the cost differential has been in the price of the VME interface silicon, which currently costs about $400 apiece. Our new device will not only boost VMEbus performance by a factor of 10, but cut the price by 70 percent thereby putting VMEbus on a price par with CompactPCI."
