🚀 go-pugleaf

>Also, after they relinquish the CPU, they also expect to recieve a message 
>from the OS, and certain data structures in the app and system are restored.
>

> If this is the reason then some system designer must have their head
> wedged.  Wouldn't it be obvious to continue providing for the app's
> volunary giving up time?  This is certainly what I'm used to from an OS --
> if it isn't true in Unix I may find myself doing some Linux kernel hacking
> sooner than I expected.
> 
> Sphere.

In article <417cks$ndh@sundog.tiac.net> Jim Williams, williams@tiac.net
writes:
>>C.mmp was an early multiprocessor project at CMU, using PDP-11's and a
>>crossbar switch to memory and I/O modules.  One of the novel features
>>of this system (at the time) was that it was an MIMD (Multiple
>>Instruction, Multiple Data) machine which symmetric (the "S" in SMP),
>>and the OS was shared across all of the processors in the system.
>>
>
>Thanks.  It sounds like SMP isn't all that much of a big deal --
>a network of micros inside my machine might even be better.  Too
>many synchronization issues...
>
>Sphere.

Once upon a time you were telling us the MacOS designers were fools for 
not including VM and pre-emptive multi-tasking in the system. 
In the above postings you reveal a complete lack of understanding of the 
issues related to pre-emption. The difference between a UNIX style OS and 
a Mac style OS is that the UNIX OS never revelealed any global state to 
apps to play around with. If it had, that UNIX OS would also have to jump 
through hoops to try to add pre-emptive multi-tasking to the system.

In fact, this exact same issue DOES arise in a UNIX system. You cannot, 
(in the UNIX systems I know of) make X-windows calls from an interrupt 
handler. Instead you have to set flags and then poll those flags in your 
main X event loop. The reason is exactly the same as what's going on with 
the mac---the X server serializes itself at the process level and if 
interupted while in the middle of a call, global state is corrupted 
and X for the system is frozen.

Next consider that Netware, generally considered the speed standard which 
other server OS's try to reach, runs with neither Protected Memory nor 
Preemtive Multi-tasking. This was (and remains) a deliberate decision.
These PMs are not free, and in some situations the code involved has a 
better idea of when to relinquish the CPU than does the scheduler.

Finally, I have to appreciate the candor of your comment on VM.
So when you chosen OS doesn't do SMP because there are "too many 
synchronization issues" that's OK, but when the MacOS doesn't do
Preemtive Multi-tasking because e are "too many synchronization issues" 
that's not OK. I guess the important thing was that you judged 
each case on the technical issues not on the vague basis that you 
understand very little of the issues involved but will nonetheless stand 
by UNIX come hell or high water because that's your team.

Maynard