Le 24/09/2013 09:05, Pierre Labastie a écrit : > Le 23/09/2013 23:30, Bruce Dubbs a écrit : >> Pierre Labastie wrote: >>> Le 23/09/2013 18:10, Bruce Dubbs a écrit : >>> >>>> On the pdf of the above, it measured KVM I/O at about 60% of the host. >>>> See pages 56-58. That seems to be consistent with my results. >>>> >>>> >>> It is really amazing, because for the SBU benchmark, I consistently get not >>> more than >>> a 10 % increase of the timings with qemu. I tried also Arch Ubuntu on the >>> VM, >>> (everything -j1, with or without 'cache=writeback', on a qcow2 format. >>> >>> On this host (core i5), I get 158 (+-2) s on the host and 170 (+-5) s on >>> the WM. >>> Both with -j1 of course. >>> >>> Here is the qemu command I use: >>> >>> /usr/bin/qemu-system-x86_64 \ >>> -monitor stdio \ >>> -smp 4 -cpu Nehalem \ >>> -soundhw ac97 \ >>> -k fr \ >>> -enable-kvm \ >>> -m 3915 \ >>> -drive file=/mnt/virtualfs/aqemu/blfs.qcow2,cache=writeback \ >>> -cdrom /mnt/virtualfs/aqemu/grub-img.iso \ >>> -boot order=c,menu=on \ >>> -net nic,vlan=0,model=e1000 -net user,vlan=0,hostfwd=tcp::2222-10.0.2.9:22 \ >>> -name "lfs" $* >> Of the options above the only thing I can see that might affect things >> are the line -smp 4 -cpu Nehalem. My system is: >> >> Intel(R) Core(TM)2 Duo CPU E8400 @3.00GHz >> >> so I can't specify -cpu Nehalem >> >> What do you get if you specify '-cpu host'? >> >> I can try specifying -smp 2 and still use make -j1 to see if that makes >> a difference. >> >> I will wait a bit because the system is doing a full LFS build (all >> tests) right now. It's chugging away at the Chapter 6 gcc tests right >> now and the automake tests always take a while. >> >> As a side note, the boot time for my system is very fast: about 2 >> seconds until starting the boot scripts and another 2 seconds to >> complete them. >> >> -- Bruce > I sometimes use -cpu core2duo and have not noticed any significant > difference. > > I would suspect that the -enable-kvm is the clue. > > I have tried without -enable-kvm, and as expected it becomes very slow. > But thinking more about it, I would guess that -machine accel=kvm only > enables acceleration for the cpu, while -enable-kvm "Enable[s] KVM full > virtualization support" (citing the man page), which might include I/O. > > No time for testing now. > > Pierre > Tested. No change... Sorry for the noise.
Pierre -- http://linuxfromscratch.org/mailman/listinfo/lfs-dev FAQ: http://www.linuxfromscratch.org/faq/ Unsubscribe: See the above information page
