On Wed, Jun 15, 2016 at 01:16:42PM +0200, Kevin Wolf wrote: > linux-aio uses a BH in order to make sure that the remaining completions > are processed even in nested event loops of completion callbacks in > order to avoid deadlocks. > > There is no need, however, to have the BH overhead for the first call > into qemu_laio_completion_bh() or after all pending completions have > already been processed. Therefore, this patch calls directly into > qemu_laio_completion_bh() in qemu_laio_completion_cb() and cancels > the BH after qemu_laio_completion_bh() has processed all pending > completions. > > Signed-off-by: Kevin Wolf <[email protected]> > --- > block/linux-aio.c | 4 +++- > 1 file changed, 3 insertions(+), 1 deletion(-)
I ran 4 x iodepth=16 random 4KB read I/O benchmarks. There might be an
improvement but it's within the error margin. My benchmarking setup can
be noisy...
Anyway, this patch doesn't hurt performance. Guest and host are RHEL 7.2.
$ ./analyze.py runs/
Name IOPS Error
linux-aio-bh-optimizations-ccb9dc1 12942616.0 ± 16.83%
linux-aio-bh-optimizations-ccb9dc1-2 13833110.4 ± 4.74%
linux-aio-bh-optimizations-off-23b0d9f 13303981.4 ± 2.21%
qemu-system-x86_64 -pidfile qemu.pid -daemonize \
-machine accel=kvm -cpu host \
-smp 4 -m 1024 \
-netdev user,id=netdev0,hostfwd=tcp::2222-:22 \
-object iothread,id=iothread0 \
-device virtio-net-pci,netdev=netdev0 \
-drive
if=none,id=drive0,file=/var/lib/libvirt/images/test.img,format=raw,aio=native,cache=none
\
-device virtio-blk-pci,drive=drive0 \
-drive
if=none,id=drive1,file=/dev/nullb0,format=raw,aio=native,cache=none \
-device virtio-blk-pci,drive=drive1 \
-display none
$ cat fio.job
[global]
filename=/dev/vdb
ioengine=libaio
direct=1
runtime=60
ramp_time=5
gtod_reduce=1
[job1]
numjobs=4
iodepth=16
rw=randread
bs=4K
Stefan
signature.asc
Description: PGP signature
