[Info-vax] Shared Stream IO (SSIO) beta kit available for OpenVMS

[Johnny Billquist]

On 2013-01-29 00:01, Stephen Hoffman wrote:
> On 2013-01-28 21:43:22 +0000, Johnny Billquist said:
>
>> On 2013-01-28 14:31, Stephen Hoffman wrote:
>>> Could fork() be implemented?  Sure.  But that means either massive
>>> complexity and untenable overhead, or ripping up the kernel.  Or both.
>>
>> copy on write is not a semantic of fork. It's just a very nice
>> optimization.
>
> With the inverse of "a very nice optimization" being "untenable
> overhead".  Lacking copy on write means spending a whole lot of time and
> processor effort replicating the contents of process address space
> around.  Sure.  Watch your cores all spin up with buffer copies,
> transferring the contents of process memory through a window and into
> the new process.   An effort which would often be followed immediately
> by tearing everything down, for the percentage of fork() calls that
> don't use that feature, too.

Indeed. A lot of copying will happen if when you don't have copy on 
write. But that would not by any means mean that a fork wouldn't work. 
Trying to behave like unix is already something fraught with inefficiencies.
Unix itself survived without copy on write for a long time, since that 
wasn't available in there as well. That's why you had inventions like 
vfork...

I'm not saying that it would be very nice or efficient, just that 
lacking copy on write is not something that in any way would prevent VMS 
from having a fork. File handling on the other hand... The whole concept 
of how that is handled in Unix is totally foreign to VMS.

	Johnny

-- 
Johnny Billquist                  || "I'm on a bus
                                   ||  on a psychedelic trip
email: bqt at softjar.se             ||  Reading murder books
pdp is alive!                     ||  tryin' to stay hip" - B. Idol