Kernel threads - so what?

Posted Aug 27, 2009 20:29 UTC (Thu) by quotemstr (subscriber, #45331)
Parent article: In brief

Err, so what if there are over 500 kernel threads? What makes 500 kernel threads "too many"? If there's a problem here, it's that the system doesn't scale to having thousands of processes, not that there are "too many" processes. A process is a fundamentally useful abstraction that a kernel would support well. The solution to poor performance isn't to hack around the problem.

to post comments

Kernel threads - so what?

Posted Aug 28, 2009 20:19 UTC (Fri) by oak (guest, #2786) [Link] (2 responses)

Threads obviously have an overhead (scheduling, task struct etc). The
less overhead you have, the better it scales.

Besides, it clutters your "ps" output. Having more kernel threads running
than the actual user-space processes for which they do stuff, sounds
wrong.

Kernel threads - so what?

Posted Aug 28, 2009 20:49 UTC (Fri) by dlang (guest, #313) [Link] (1 responses)

computers can easily track several thousand inactive threads with very little overhead, but the sysadmin trying to examine ps or top output to figure out what's happening has a much lower threshold

Kernel threads - so what?

Posted Sep 8, 2009 9:04 UTC (Tue) by jlokier (guest, #52227) [Link]

Obvious solution is to hide kernel threads in ps/pstree/top.

It would be nice if the kernel threads had lazy stacks too, to save a bit of memory.