> yuri mironoff <yuri@rgti.com> wrote:
> > Not quite, the returning AST under VMS interrupts the "parent" process
> > at a previously provided handler function. I'm not sure thats possible
> > under the current thread inplementations.
I worked for years under VMS and have developped lots of application
that used asts and tries to deal with severall contexts simultaneously.
Programmers are generally happy of this mechanism and I was too.
But you have to pay the overhead that each request must be queued, memory
allocated from various pools and deal with lots of quota and limits that
make tuning painfull.
I have developped similar application under variouses unices.
Obviously I must deal with select(), signals and processes.
The result is that such applications are easier to tune under unix
and they have far less overhead.
I have developped too some software under NT.
Just for your information:
The CPU load of 1 call to a system service under NT seems
to be about 5 to 10 times the CPU load to the equivalent
service under Linux. Must we have a so brain-deaded system in order
to have "a la VMS" system services.
In my opinion, using an OS with apparently clever system services but
so heavy that we donnot want to use system services in order to have
performance is loose.
> exit(), SIGCHLD, waitpid() should provide the necessary machinery.
> The pthread API is a bit deficient, however (but that's a library issue).
Gerard.