dataplane: submit I/O as a batch
commitdd67c1d7e75151e2c058ccdd2162643074357442
authorMing Lei <ming.lei@canonical.com>
Fri, 4 Jul 2014 10:04:35 +0000 (4 18:04 +0800)
committerStefan Hajnoczi <stefanha@redhat.com>
Mon, 7 Jul 2014 09:05:17 +0000 (7 11:05 +0200)
tree31e040b1efae71738f98ab05ba4e0236e2aed9d8
parent1b3abdcccf18d98c3952b41be0bc1db3ef6009dd
dataplane: submit I/O as a batch

Before commit 580b6b2aa2(dataplane: use the QEMU block
layer for I/O), dataplane for virtio-blk submits block
I/O as a batch.

This commit 580b6b2aa2 replaces the custom linux AIO
implementation(including submit I/O as a batch) with QEMU
block layer, but this commit causes ~40% throughput regression
on virtio-blk performance, and removing submitting I/O
as a batch is one of the causes.

This patch applies the newly introduced bdrv_io_plug() and
bdrv_io_unplug() interfaces to support submitting I/O
at batch for Qemu block layer, and in my test, the change
can improve throughput by ~30% with 'aio=native'.

Following my fio test script:

[global]
direct=1
size=4G
bsrange=4k-4k
timeout=40
numjobs=4
ioengine=libaio
iodepth=64
filename=/dev/vdc
group_reporting=1

[f]
rw=randread

Result on one of my small machine(host: x86_64, 2cores, 4thread, guest: 4cores):
- qemu master: 65K IOPS
- qemu master with these patches: 92K IOPS
- 2.0.0 release(dataplane using custom linux aio): 104K IOPS

Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
hw/block/dataplane/virtio-blk.c