Vincent Diepeveen (diep@xs4all.nl) on 5/16/07 wrote:
---------------------------
>You just reserve 32MB ram for shared memory, any allocation bigger than that needs
>root permission. Very ugly from user viewpoint.

erm the maximum size of any shared memory segment and the total amount of shared memory is a tunable kernel parameter via /proc.

>Not many systems nowadays have just one core.

Almost all installed systems have one core. Most new systems have more than one core.

>you just see how every few seconds the processes swap from processor to processor,
>based upon the load of the processor.
>
>Very ugly.

No it's by design. It is a fundamentally hard problem to design a scheduler that can know if it is better to wait and restart a thread on the same core or to migrate it.

If you know better than the scheduler what's going on as you have some knowledge that you are running one thread per CPU for long periods of time then you can the sched_set_affinity() calls to tell tell the scheduler which core it is allowed to schedule a thread on.

>Just my own private opinion...

Most of these things you are complaining about which may appear to be simple and obvious on the surface are either fundamentally hard computer science problems or problems which are hard due to grotty underlying architecture that the OS is shielding you from.

If you want something which achieves your goals then read up about the problem and propose what algorithms could be used to solve the problem.

Failing that dispense with the OS entirely and just run DIEP as it's own OS...