[I resend my message because MLs have refused the first one in HTML] On 28/05/2015 07:17, Paul Mackerras wrote: > This patch series provides a way to use more of the capacity of each > processor core when running guests configured with threads=1, 2 or 4 > on a POWER8 host with HV KVM, without having to change the static > micro-threading (the official name for split-core) mode for the whole > machine. The problem with setting the machine to static 2-way or > 4-way micro-threading mode is that (a) then you can't run guests with > threads=8 and (b) selecting the right mode can be tricky and requires > knowledge of what guests you will be running. > > Instead, with these two patches, we can now run more than one virtual > core (vcore) on a given physical core if possible, and if that means > we need to switch the core to 2-way or 4-way micro-threading mode, > then we do that on entry to the guests and switch back to whole-core > mode on exit (and we only switch the one core, not the whole machine). > The core mode switching is only done if the machine is in static > whole-core mode. > > All of this only comes into effect when a core is over-committed. > When the machine is lightly loaded everything operates the same with > these patches as without. Only when some core has a vcore that is > able to run while there is also another vcore that was wanting to run > on that core but got preempted does the logic kick in to try to run > both vcores at once. > > Paul. > --- > > arch/powerpc/include/asm/kvm_book3s_asm.h | 20 + > arch/powerpc/include/asm/kvm_host.h | 22 +- > arch/powerpc/kernel/asm-offsets.c | 9 + > arch/powerpc/kvm/book3s_hv.c | 648 ++++++++++++++++++++++++++---- > arch/powerpc/kvm/book3s_hv_builtin.c | 32 +- > arch/powerpc/kvm/book3s_hv_rm_xics.c | 4 +- > arch/powerpc/kvm/book3s_hv_rmhandlers.S | 111 ++++- > 7 files changed, 740 insertions(+), 106 deletions(-) Tested-by: Laurent Vivier <lvivier@xxxxxxxxxx> Performance is better, but Paul could you explain why it is better if I disable dynamic micro-threading ? Did I miss something ? My test system is an IBM Power S822L. I run two guests with 8 vCPUs (-smp 8,sockets=8,cores=1,threads=1) both attached on the same core (with pinning option of virt-manager). Then, I measure the time needed to compile a kernel in parallel in both guests with "make -j 16". My kernel without micro-threading: real 37m23.424s real 37m24.959s user 167m31.474s user 165m44.142s sys 113m26.195s sys 113m45.072s With micro-threading patches (PATCH 1+2): target_smt_mode 0 [in fact It was 8 here, but it should behave like 0, as it is > max threads/sub-core] dynamic_mt_modes 6 real 32m13.338s real 32m26.652s user 139m21.181s user 140m20.994s sys 77m35.339s sys 78m16.599s It's better, but if I disable dynamic micro-threading (but PATCH 1+2): target_smt_mode 0 dynamic_mt_modes 0 real 30m49.100s real 30m48.161s user 144m22.989s user 142m53.886s sys 65m4.942s sys 66m8.159s it's even better. without dynamic micro-threading patch (with PATCH1 but not PATCH2): target_smt_mode 0 real 33m57.279s real 34m19.524s user 158m43.064s user 156m19.863s sys 74m25.442s sys 76m42.994s Laurent -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
