Hi Folk! On Fri, Mar 21, 2003 at 09:53:18PM +0530, Gagan Grewal wrote:
It seems the mail to remove the confusion didn't get through to the list. Am attaching it again with this mail. Regards, Gagan.
Hi Dhruv! Dhruv Gami wrote: >Could you please elaborate on whatever u said ? i dont know about the >others, but all this went over my head, and now im curious to know what >was being talked about :) > Well ... I'll try to simplify LWPs or as Supreet very correctly said ... clones on Linux ... are a way of implementing schedulable executing enitites which are not processes but execute as a part of a process. If you are thinking 'threads' then you are right :) There are three types of scheduleable entities which can be created: 1. kernel threads : run in kernel space, created using the function kernel_thread. The project I proposed doesn't touch this as of now, but if the need arises ... we might have to use this function. 2. clones : run in user space, created using the clone() lib routine, which calls sys_clone() function from inside the kernel. This is what I want to focus on. The advantage being that individual functions can be scheduled independently. ( If you have objections with the word 'scheduled', I'll comply ... :) ) 3. processes : run in user space, created using fork() lib routine, which calls sys_fork() from inside the kernel. Not interested in processes directly. A small summary: fork() , clone() are library routines wrapped around sys_fork() and sys_clone() kernel functions. Both these functions are written on top of do_fork() kernel function which has argument 'clone_flags' besides two other related to the stack (and another one related to context) (I could go to lengths describing the entire stuff, but wanna keep it simple) So simply ... the fork() routine ... or the creation of an entire process is just a special case of a clone() creation where the clone_flags force the clone to turn into a full process. So basically, clones are available as/in-place-of LWPs, what needs to be seen wethet a solution for C++ programmers can be found or not. Now what I want to do is to have a set of classes written in C++ which allows us to instantiate objects which execute functions independently ... as in concurrency. i.e. to have a clone() call embedded in the class constructor or filtered elsewhere but instantiation of object should result in a clone or thread. The reason should be clear if you yourself try encapsulating fork() inside a class, 'exec'ing and then go ahead and try implementing synchronization also. Can be done ... no doubt. But would be a lot easier if it were done using clone() and if all this were wrapped up neatly in C++ classes. (I managed to implement ONE-MANY client-server skeleton using Semaphores but ... it was done in C ... I can't imagine doing it in C++ ... OK ... it can be done but if one is to implement code-reuseability ... he'll have to work real hard) One of the solutions is infact wrapping up POSIX threads inside classes. (But is that fun ? besides ... adding another library to be linked :( ) There are other solutions, but most of them state ... "a set of C++ wrappers is under current developement" I want to start by wrapping up clone() call itself and simplify the stuff into classes. (Again ... I request that someone knows of this thing being done already plz let me know ... I tried googling but results were as mentioned above) So ... if there are more doubts ... plz write in ... that'll help me too. Regards, Gagan.
