Am 09.10.2015 um 09:49 hat Paolo Bonzini geschrieben: > On 08/10/2015 11:41, Kevin Wolf wrote: > > Am 08.10.2015 um 10:54 hat Stefan Hajnoczi geschrieben: > >> On Thu, Oct 01, 2015 at 01:04:39PM +0200, Paolo Bonzini wrote: > >>> Simplify memory allocation by sticking with a single API. GSlice > >>> is not that fast anyway (tcmalloc/jemalloc are better). > >>> > >>> Signed-off-by: Paolo Bonzini <pbonz...@redhat.com> > >>> --- > >>> block/io.c | 4 ++-- > >>> block/mirror.c | 4 ++-- > >>> block/raw-posix.c | 8 ++++---- > >>> block/raw-win32.c | 4 ++-- > >>> hw/block/virtio-blk.c | 4 ++-- > >>> 5 files changed, 12 insertions(+), 12 deletions(-) > >> > >> Thanks, applied to my block tree: > >> https://github.com/stefanha/qemu/commits/block > > > > Has someone benchmarked this before applying? Just claiming "wasn't fast > > anyway" doesn't generally seem sufficient for changes to the I/O path. > > I did it about six months ago. Sorry for not digging up the results > when posting: > > baseline: 193 kiops > tcmalloc: 202 kiops > tcmalloc + G_SLICE=always-malloc: 210 kiops
Thanks. Do you have numbers for g_malloc + G_SLICE=always-malloc, too? Because I think that's our new default after this patch. tcmalloc must still be enabled manually. Kevin
