Re: [Corelinux-public] Re: Class Library and Framework Requirements forum msg #7915 (Threads)

[Jacob C. <jac...@ho...>]

>"Frank V. Castellucci" wrote:

[snip]
>I think we have two options:
>
>One-to-One : Uses the Linux "clone" api which is more efficient than
>fork. I perceived that clone, even though it calls it a process in the
>'ps' display, is still running in the address space of the parent and
>they share the memory space. Thread n modifies something and it is
>visible to the other threads or main process thread.
>
>Many-to-Many : Combo of "clone" and a homegrown scheduler that takes its
>que from the kernel.
>
>I believe the latter to be an expensive (time, resources, acceptance)
>proposition but attractive none the less.


for performance reasons i presume? (referring to attractiveness)

we may be able to mitigate some of the time necessary for the many-to-many
implementation by taking a look at the scheduling code for GNU Pth.  I think
it is
under the GPL so we can't cut'n'paste but taking a look i think would be
useful.


>If we think foward to the frameworks endeavor, it is reasonable to
>construct a Thread or Process framework that can be used to serve up
>default libcorelinux++ Thread (one-to-one) or whatever the framework
>extension desires (many to many, etc.).


i agree.  this would leave the door open for a many-to-many implementation
with a readily useable one-to-one implementation.