[Bug 254639] Fork syscall performance

bugzilla-noreply at freebsd.org bugzilla-noreply at freebsd.org
Mon Mar 29 15:26:17 UTC 2021


https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=254639

Mark Johnston <markj at FreeBSD.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |markj at FreeBSD.org

--- Comment #1 from Mark Johnston <markj at FreeBSD.org> ---
How much RAM do you have?  Hopefully more than 100GB.

I suspect that we're spending a lot of time in pmap_copy(), which copies page
tables into the child process.  It only copies referenced PTEs, but in this
case we'll copy all of the page tables backing some_str.

You can try something like this to confirm:

# dtrace -q -n 'fbt::pmap_copy:entry /pid == $target/{self->ts = timestamp}
fbt::pmap_copy:return /self->ts/{@ = sum(timestamp - self->ts); self->ts = 0;}
tick-1s {printa("spent %@dns copying\n", @); clear(@);}' -c "python test.py"

-- 
You are receiving this mail because:
You are the assignee for the bug.


More information about the freebsd-bugs mailing list