On Sun, 3 Oct 2004 08:47:07 +0545, manish regmi <regmi.manish@xxxxxxxxx> wrote: > If your goal is to implement another scheduling algorithm. I think > there is no reason for changing the fork system call. Did you read his original message? He said that he needs to pass in deadlines (and likely other parameters which differ from those of the standard Linux scheduling algorithm). > The better way is to change few things in sched.c. It is recommended > to add a new scheduling policy NOT change the existing ones. I agree it's worth adding a new scheduling policy, allow it to be set with setparam and then come up with a way to pass in the additional parameters to the kernel - the trouble is that he probably wants to set up deadlines at process creation rather than any time thereafter (what would be the point otherwise?) so he needs to have a mechanism to pass these in to the kernel at clone-time. I'll think about the best way to implement this - a quickie hack would be to modify clone with a new flag so we don't start executing the new process until a later system call passes in a some data but that's really ugly and we want this to be smooth and fast. Unless I'm missing something he'll probably end up with changing/add syscalls. I've got the POSIX standard here somewhere (and so do you if you're in academia and have some equivalent of the ATHENS system we have in the UK for obtaining this stuff free) and there's probably some stuff in .4 which should guide what to do here. Will look when I'm not planning to go out through the rain in search of coffee and OxLUG :-). Jon. -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/
