[COLUG] Presentation Follow-up: Linux Kernel Scheduling

[Travis Sidelinger]

After doing some research on the clock timer (PIT - programmable 
interval timer), it is indeed a non-maskable interrupt (Interrupt 0). 
It's interval is stored in the kernel value tick_nsec, which is about 
1msec.  The kernel macro HZ will return its value, and it can be 
calculated from watching /proc/interrupts.  The scheduler_tick() 
function is called on every clock timer "tick", which performs basic 
scheduler accounting.

I was wrong when I said that domains can be controlled at boot time.  I 
was confusing that with a boot option to force cpu's outside the 
schedulers control.  The isolcpus= option can be used to isolate a cpu, 
then later the taskset tool could be used to place processes on that 
isolated cpu.

The formation of domains appear to be hardware dependent.  Also, 
re-balancing can be performed across processor domains.  Though, the 
domains as whole entities must qualify for re-balancing.  This is where 
on large systems, setting affinity may become important for a group of 
processes.

My wiki page for this material is located here:
   http://www.ilive4unix.net/doku.php/notes/unix/linux/scheduler

I was not planning on posting the presentation ods file, unless someone 
really wants it.  Wiki's are much easier to maintain.

Yet me know if you have any other follow up items.


Travis Sidelinger