Blade server progress report

Matthew Dillon dillon at apollo.backplane.com
Wed Aug 7 23:12:05 PDT 2013


    We received our blade server.  It isn't installed in the colo facility
    yet but that will hopefully happen in the next week or two.  Right now
    we are moving services from the discrete machines to the blades.

    These haswell blades are blasted fast and will idle down to 18W each.
    I am very impressed.

    We had some fun with pkgbox32.  Since pkgbox32 is running 32-bit it
    would have been a waste to dedicate a 16GB blade to it, so we actually
    installed linux on that particular blade and we are running the
    DragonFly pkgbox32 as a qemu KVM instance (with virtio no-less).
    Now, of course, the Haswell xeon is two generations ahead of the Phenom II
    that the old pkgbox32 was running on.  But still, both are clocked at
    3.4 GHz and I set ncpus to 4 in the VM to match the Phenom II, then
    ran a buildworld -j 8 test.  The qemu instance is 40% *faster* than
    the old pkgbox32 was.

    I am impressed at how far hardware virtualization technology has
    progressed.

    In anycase, pkgbox32 and pkgbox64 have been moved, and the main dev
    box (leaf) and the main repo box (crater) will be moved to blades
    tomorrow (Thursday).  Avalon and kronos will be moved after the
    unit is installed in the colo facility.

    There may be some hicups as we do this.  Best thing to do is to get
    onto irc (efnet) in #dragonflybsd and report issues there.  Otherwise
    bugs@ mail works.

					-Matt
					Matthew Dillon 
					<dillon at backplane.com>


More information about the Users mailing list