Path: utzoo!attcan!uunet!lll-winken!ames!ncar!boulder!tramp!hassell From: hassell@tramp.Colorado.EDU (Christopher Hassell) Newsgroups: comp.sys.apple Subject: Re: Is anyone here interested in the "Future of Apple //?" Message-ID: <5833@boulder.Colorado.EDU> Date: 15 Jan 89 09:39:04 GMT References: <5678@boulder.Colorado.EDU> <1209@umbio.MIAMI.EDU> <9323@smoke.BRL.MIL> <5747@boulder.Colorado.EDU> <788@convex.UUCP> Sender: news@boulder.Colorado.EDU Reply-To: hassell@tramp.Colorado.EDU (Christopher Hassell) Organization: (Let's see, I'm positive .. I've got ... ) Lines: 92 buyse@concave.UUCP (Russell C. Buyse) writes: #Christopher Hassell writes: #>Doug Gwyn (VLD/VMB) writes: #>#It may come as a surprise to people who haven't had experience using #>#parallel processing, but it is hard to exploit multiple concurrent #>#processors in a reliable, controlled manner. For a handful of CPUs, #>#probably the best way to use them is in support of a mutiprocessing #>#environment such as UNIX. #> #>That school of thought has TOO much "common knowledge" for my tastes, sorry. #>Very coarse grained parallelism is a perfect application for micros, I #>believe. # #I think that you are tossing away the issues of multi-processing #too quickly. Devising architectures, system software, and development #tools is a much more complex task when you throw parallelism into the #formula (perhaps a reference to comp.parallel is in order?). If #parallelism had so many built-in advantages with micros, there would be a #good deal more of them available that do it. IBM has shown us that innovation is not necessarily the best business strategy. The processors are not necessarily equivalent in environment, even though that would be nice. Architectures would only require things like locking write-access to I/O and multi-porting RAM in certain places in certain ways. Otherwise very little would need to be done other than package two machines into one. It is their overlap that is powerful but their separateness also helps to simplify things. System software is dealing with two equivalent processors but not two that do the same tasks or work in *completely* similar manners (w/out inherent knowledge regarding which one runs what stuff). With things like that there is very little to worry about in system software except (again) things like write-locking certain stuff, and even then it is *coarse* parallelism so separate memory is very likely. #Parallelism is not required for the next Apple II. What is required is #much faster scalar execution and enhanced graphics capabilities. These #are both well-known quantities that can be produced by Apple Computer. Well, nothing is "required" for the next Apple II. I feel that given the amiga as a competitor, the Apple will still not live up to expectations that are requiring a lot more because of the amiga's example regarding gee-whiz hardware niceties. #The best way that multiple processors might be utilized would be as it is #being done in many micros-- by using processors for individual functions, #such as for the keyboard, etc, in order to free the CPU for other tasks. The advantages I speak of a very similar to the things you suggest *except* that they are all combined into the versitility of a SOFTWARE processor instead of near-hard-wired stuff. Interrupts can help distribute the raw power a second processor would produce. A *plethora* of other things, however are likely, specifically for micros. Paging types of activites are already done but simple lookups and occasional backing up of a memory image are other *very nice* possibilities. Many things regarding the user-interface these days have NO END-SYNCHRONIZATION, that is to say, nearly nothing depends on when some jobs are done. Graphics are one application that quickly comes to mind, along with pre-loading of sound info before it is needed. Other very nice possibilities are that of being able to have a smart end handling things on a network (notification when printing is done, when mail arrives etc...). Many many things can be done with I/O that is not interactive. Studies have shown that ONLY ONE background task is usually required at any time in a working session, with buffers or not. So there they are: Extended Smart I/O stuff, Memory handling, Non-critical jobs, and general versitility are all possible WITHOUT bringing stuff like nice safe multiprocessing in (usable ONLY after multi-tasking has been established). Note things that are not provided in this idea are things like n-processor parallelism and complete switchability of processor environments. I still think the benefits would produce a far more useful computer than any based on vaporMhz and most of those on the market today. Please don't anyone consider any of this to be aimed at any people or at any general principles, except that parallelism has been treated like plutonium, and no one has figured out that it is as safe as sunlight if used with as much versitility as is possible. #-russ. #UUCP: {uiucdcs,sun,uunet,harvard,killer,usenix}!convex!buyse # --or-- # buyse@convex.COM ### C>H> ###