Another neat question Jim Privat asked:
>(2) Will the trend be that more and more of the smarts will be built into the hardware, eventually obviating the need for software on the chip?
Remember the fall of 1994 when Intel announced its NIS plans at the fall COMDEX, which would enable all DSP-type devices to be integrated in software within the CPU? Add-on board manufacturers’ stock plummeted. I remember, because I owned CUBE, and it got tanked on the news. Can you imagine software replacing an MPEG-2 encoder? Well…maybe it can.
The fact is that hardware and software are in an unending battle for total performance vs. total cost. You seem to think that, in the end, hardware might win. The above exemplifies a possible counter-trend. Actually, this is a battle I don’t want to on either side.
The main point is there is nothing I see on the horizon that obsoletes the notion of an RTOS per se. Could it be implemented in hardware? Probably, but only at a major cost of lost flexibility, with probably extremely small gain in performance.
Nevertheless, suppose that is exactly what TI wants to do with their 125 million transistors in a chip the size of a thumb nail. Where do they get the logic flow for the RTOS? For the exact same reasons that Intel chose VxWorks (IxWorks) for the i960 RP chip, they would likely choose VxWorks, or some other commercial RTOS, to cast into silicon. The commercial vendor would get a royality either way.
Probably the most challenging problem commercial RTOS vendors have is keeping up with the plethora of new processors, each in many flavors. To hamstring them by adding that they add a "cast in silicon" step would not be welcomed in the foreseeable future. Don’t forget, many applications benefit from the ability of EPROM upgrades delivered telephonically and other ways, because they are implemented in software.
Thus, the on-going battle between the hardware and software does not obsolete the RTOS vendors, no matter who wins. There is no paradigm shift on the horizon that obviates the RTOS product, whether it is ultimately delivered in software or hardware.
PS. At the H&Q Tech Show, Mentor Graphics attempted to describe a new direction in which they will be including OS concepts in the design of new chips. They implied this is why they bought Microtech Research, and that Microtech Research will be focusing on this aspect of chip design. However, exactly what this means, how they intend to pursue it technically, and whether it is supposed to orient the final chip product toward associated flavors of VRTX, or even contain VRTS was unclear to me.
PSS. READ THIS CAREFULLY, IT IS VERY IMPORTANT. IxWorks deployment in i960 RP was the first time an RTOS was so closely coupled with a processor, and is stirring the industry. Generally, one expects a processor vendor to want to de-couple from the RTOS vendor in order to be considered by as many parties as possible. Intel broke the mold with its I2O implementation. Maybe the next generation will cast the RTOS in silicon, or maybe that is exactly what Mentor Graphics is trying to do. WIND still gets its growing share of RTOS business, as will MWAR and possibly others. (See the next discussion on market share for more about how this plays out.)
Allen, |
