[gsoc] Virtio Block device benchmarks

[Stéphanie Ouillon <stephanie@xxxxxxxxx>]

Hello,

I intended to test the performances of the virtio block device driver 
(the one which is on gitorious 
https://gitorious.org/virtio-drivers/virtio-drivers/blobs/master/blk/virtio-blk.c).

Virtualization environnement:

host: QEMU emulator version 0.14.0 (qemu-kvm-devel), on Proxmox VE 1.8
guest: DragonFly BSD i386 v2.10.1, RAM 1024

Filesystem: Hammer
emulated IDE hard disk - 20GB
virtio hard disk -10GB
(format of the image : qcow2)


To mount the virtio disk, I did:

    # newfs_hammer -f -L disk-virtio /dev/vdb0s0
    # mount_hammer /dev/vdb0s0 my-mount-point


I made some tests

- by running dd commands like :
# dd if=/dev/zero of=/root/my-mount-point/file1 bs=4k count=10
# dd if=/dev/zero of=/root/file2 bs=4k count=10
and changing the value of the bs and the count fields.

- via bonnie++ tool

The first point is that an average would have to be calculated so that 
the results have a real meaning because they fluctuate "a lot" from a 
test to another with the same arguments. Sometimes it is faster on the 
virtio disk, sometimes it is not.
What could be a convincing way to benchmark it ? Tim Bisson ran dd 
parallel threads, is it the only way to ?


But the main point is that  the VM often crashes during the test and I 
have to reboot (I can't connect anymore via SSH).
It sometimes crashes during a dd test.
It always crashes during bonnie++ tests. The first test performed by 
this tool is to write repetitively in a file with the standard stdio 
macro putc().

I run the command :
bonnie++ -u root -d /root/my-mount-point/ -r 1024
where -r specifies the size of RAM

The test is lauched at 22:21:38. This is what appears in /var/log/messages :

Jun  3 22:17:58  kernel: HAMMER(disk-virtio) recovery range 
3000000000018368-3000000000024830
Jun  3 22:17:58  kernel: HAMMER(disk-virtio) recovery nexto 
300000000001ffe8 endseqno=000fc247
Jun  3 22:17:58  kernel: vringdescnext
Jun  3 22:17:58  last message repeated 8 times
Jun  3 22:17:58  kernel: HAMMER(disk-virtio) recovery undo  
3000000000018368-3000000000024830 (50376 bytes)(RW)
Jun  3 22:17:58  kernel: vringdescnext
Jun  3 22:17:58  last message repeated 86 times
Jun  3 22:17:58  kernel: HAMMER(disk-virtio) recovery complete
Jun  3 22:17:58  kernel: vringdescnext
Jun  3 22:18:17  last message repeated 14 times
Jun  3 22:21:38  kernel: vringdescnext
Jun  3 22:21:38  last message repeated 9 times
Jun  3 22:21:38  kernel: here
Jun  3 22:21:38  kernel: vringdescnext
Jun  3 22:21:38  last message repeated 104 times
Jun  3 22:21:38  kernel: Bad enqueue_reserve
Jun  3 22:21:38  kernel: vringdescnext
Jun  3 22:21:38  last message repeated 279 times
Jun  3 22:21:38  kernel: here
Jun  3 22:21:38  kernel: Bad enqueue_reserve
Jun  3 22:21:38  kernel: vringdescnext
Jun  3 22:21:39  last message repeated 116 times
Jun  3 22:21:39  kernel: here
Jun  3 22:21:39  kernel: Bad enqueue_reserve
Jun  3 22:21:39  kernel: vringdescnext
Jun  3 22:21:39  last message repeated 100 times
Jun  3 22:21:39  kernel: here
Jun  3 22:21:39  kernel: Bad enqueue_reserve
Jun  3 22:21:39  kernel: vringdescnext
Jun  3 22:21:39  last message repeated 89 times
Jun  3 22:21:39  kernel: here
Jun  3 22:21:39  kernel: Bad enqueue_reserve
Jun  3 22:21:39  kernel: vringdescnext


I just quickly looked through the code to see what is the problem :

This error message appears in virtio-blk.c, in the virtio_blk_execute 
function, at ligne 235.
virtio_enqueue_reserve return a bad value ( <>0 ) at ligne 535 of 
virtio.c if (qe==NULL), that is to say, that there is no free place 
anymore in the virtqueue.

I'll look at it more precisely later during the week-end, but if you 
have some ideas about it...


Stéphanie