This whole thread should be useful.
I would look for a way to limit the 'init' process to running on cpu 1.
In general case the last words should be "running on all CPUs but 1". I guess one can use some kernel parameters; irqaffinity, isolcpus look promising.
EDIT:
About isolcpus in kernel parameters documentation:
This option is the preferred way to isolate CPUs. The alternative -- manually setting the CPU mask of all tasks in the system -- can cause problems and suboptimal load balancer performance.
I did some tests. Results:
Kernel parameter
isolcpusis the one you may use to limit theinitprocess and isolate CPU(s).It seemed to me that kernel parameter
irqaffinitydid not affect isolated CPU(s). You probably don't need this parameter.When I isolated CPU0 it didn't go completely idle. Is that system or hardware requirement to do some computation with it despite isolation? – I don't know.
When I isolated CPU1 or CPU7 it was
solid asidle as a rock.
To isolate CPU1 add isolcpus=1 in your grub.cfg, like this:
linux /vmlinuz-(…all your current parameters here…)isolcpus=1
ORIGINAL ANSWER CONTINUES:
Alternatively, after the system has started, you can re-assign the affinity of all currently running processes, but that seems less elegant.
I agree it seems less elegant, yet it may work good enough (EDIT: it can cause problems and suboptimal load balancer performance, as stated above). There is quick and dirty one-liner that works in my Debian:
sudo bash -c 'cd /proc ; for i in [1-9]* ; do taskset -a -p fffffffe $i ; done' I got errors for some pids, nevertheless htop shows that one core is virtually idle after that. Of course you have to run your script later, exclude it somehow or simply reassign:
taskset -a -p 1 PID I expect you won't need to repeat said one-liner to cover new processes since CPU affinity is inherited.