Dragonfly and Hyperthreading....

Matthew Dillon dillon at apollo.backplane.com
Mon Feb 21 21:48:28 PST 2005


:On Tue, 22 Feb 2005 03:27:38 +0100, Joerg Sonnenberger wrote:
:
:> On Mon, Feb 21, 2005 at 07:05:52PM -0700, Dan Eicher wrote:
:>> I've been testing DragonFly on a dual 242 opteron Via K8T Master2-FAR mobo
:>> and everything Just Works. The only issue I've seen is when doing a
:>> buildworld it doesn't utilize the whole power of the cpus, it pretty much
:>> just tops out at 146% according to "time make -j 3 buildworld".
:>> Buildkernel can get it up to 169.8% but the same tests on FreeBSD 5.3 and
:>> current would go ~196%.
:> 
:> That's not really surprising. Buildworld does depend quite a lot on IO,
:> which isn't parallel yet.
:> 
:> Joerg
:
:I figured it was an IO bottleneck. As an aside, I also get quite a few
:"[diagnostic] cache_resolve: had to recurse on <dir>"  and 1 "Warning:
:vnode 0xda1708f8 clean/cache_resolution race detected" messages. I'm 
:playing around with the gcc34 version and am very impressed with the
:quality and stability. 
:
:Dan

    Yah, you can ignore those messages.   There was a livelock race
    between cache_inval() and cache_resolve() that was recently reported
    and fixed and the diagnostic messages are related to the preconditions
    for it.  It's fixed now so you can ignore the messages, I just wanted
    to get an idea at how often the situation occured.

    Now that the TLS support is in my current near term TODO list is:

    * Andrew Atrens vm_contig patch.

    * A buffer cache red-black tree implementation patch that I almost have
      working.

    * And the real time journal.

					-Matt
					Matthew Dillon 
					<dillon at xxxxxxxxxxxxx>





More information about the Users mailing list