SMP performance on drgonfly

Matthew Dillon dillon at
Sun May 20 13:55:11 PDT 2007

:I was hoping that Matt would reply to my questions above so I could
:try to find a more level playing field on which to compare Dragonfly
:and FreeBSD, but I'm afraid that the answer is that there currently
:are *no* kernel subsystems that are not Giant-locked, so all workloads
:will perform poorly on DragonFly (and there are no specialized tools
:for studying SMP behaviour -- certainly I couldn't find any when I
:Can someone confirm whether this is true?

    Well, you can set kern.syscall_mpsafe=1 and kern.trap_mpsafe=1
    (don't set kern.intr_mpsafe to 1, though).

    With those set, *SOME* system calls will not need the big giant lock.
    e.g. things like getuid() and dup().  But anything doing I/O will
    still get the lock.  The BGL will be pushed in a little even for
    system calls that still need it, but not by much.  All significant
    interrupts still need it, too.

    We're looking for volunteers to help lock up subsystems.  The main
    one that is just about ready is the network stack.

					Matthew Dillon 
					<dillon at>

More information about the Users mailing list