Re: [Qemu-devel] [PATCH 00/10] AHCI emulation support v2

[Alexander Graf]

On 18.11.2010, at 21:06, Ryan Harper wrote:

> Speaking of fast, do you have any numbers around ACHI vs IDE (not that I
> need any convincing that we can do better than IDE); just curious.

To test the raw link speed, I usually take a tmpfs backed sparse raw file and 
pass it to the guest. Inside the guest, I then dd the device to /dev/null with 
iflag=direct, so we don't get the guest page cache involved.

ide:

linux-uztg:~ # dd if=/dev/sdc of=/dev/null bs=10M count=300 iflag=direct
300+0 records in
300+0 records out
3145728000 bytes (3.1 GB) copied, 2.06424 s, 1.5 GB/s

ahci:

linux-uztg:~ # dd if=/dev/sda of=/dev/null bs=10M count=300 iflag=direct
300+0 records in
300+0 records out
3145728000 bytes (3.1 GB) copied, 1.40743 s, 2.2 GB/s

virtio:

linux-uztg:~ # dd if=/dev/vda of=/dev/null bs=10M count=300 iflag=direct
300+0 records in
300+0 records out
3145728000 bytes (3.1 GB) copied, 0.851306 s, 3.7 GB/s


Obviously, the main benefit of ahci over ide is that ahci can have multiple 
outstanding read/write requests at the same time. So even with the same link 
speed, you would see improvements because the backend can be partially cached 
or striped over different volumes which make simultaneous requests crucial for 
performance.


Alex