Windows crashes under heavy I/O load?

F

flypig

New Member
Sep 1, 2011
17
0
1
Hello,

I have a test system on Proxmox 2.0 Beta.

After moving a windows 2008R2 vm from promox 1.9 to 2.0 successfully.
I notice the windows startup prompted for driver for the new virtio io controller, the old virtio io driver is no longer accepted, therefore I downloaded the latest virtio-win-0.1-15.iso from fedora website.

Everything works smoothly with the new driver until I ran a IO benchmark / load test.
The name of the programe is: CrystalDiskMark 3.0.1 x64 (crystalmark.info)

hddtest1.png
This screenshot is captured seconds before it crashes. I have ran the same test 10 times and all 10 test crashed at 4k random write test.
(Hardware LSI RAID 0, Stripe size 64k)

The first 2 test: sequencial and 512k random read/write went successfully.
On the 3rd test: random 4k read test went successfully, however on the write test the blue screen of death came out.

err1.pngerr2.png

Below is the screencap of the cpu, io, mem utilization graph.

proxmox1.png

I then turn to my other test system (same RAID disk setup) which is using Proxmox 1.9, all the test went successfully without any problems.
note that on 1.9 windows is using the older virtio io driver.


It seems that the blue screen only came out under heavy IO write at 4K (since my stripe size is 64k, 4k will only write to 1 disk)
Could it be issue with the latest virtio io driver?


root@vps:~# pveversion -v
pve-manager: 2.0-4 (pve-manager/2.0/52943683)
running kernel: 2.6.32-6-pve
proxmox-ve-2.6.32: 2.0-46
pve-kernel-2.6.32-6-pve: 2.6.32-46
lvm2: 2.02.86-1pve1
clvm: 2.02.86-1pve1
corosync-pve: 1.4.1-1
openais-pve: 1.1.4-1
libqb: 0.5.1-1
redhat-cluster-pve: 3.1.7-1
pve-cluster: 1.0-7
qemu-server: 2.0-1
pve-firmware: 1.0-13
libpve-common-perl: 1.0-5
libpve-access-control: 1.0-1
libpve-storage-perl: 2.0-4
vncterm: 1.0-2
vzctl: 3.0.29-3pve1
vzdump: 1.2.6-1
vzprocps: 2.0.11-2
vzquota: 3.0.12-3
pve-qemu-kvm: 0.15.0-1
ksm-control-daemon: 1.1-1
 
Last edited:
flypig said:
...
It seems that the blue screen only came out under heavy IO write at 4K (since my stripe size is 64k, 4k will only write to 1 disk)
Could it be issue with the latest virtio io driver?
..
Click to expand...

1.9 and 2.0 uses more or less the same kernel and also the same version of KVM. can you test the latest virtio drivers also on 1.9?
 
Hi Tom,

Thank you for your prompt reply,
I guess I had figured out the problem.

The blue screen of death dissapears if I set cache: no-cache !

Earlier I was using "Write-Back" cache, thinking I could use this since my raid card supports write-back with bbu.

cache.png

After setting to no-cache, the 4k random wite test went perfectly without any problems.
However notice that the preformance degraded much compare to the earlier test with write-back cache enabled.

final.png

The harddisk cache option seems to be a new feature in proxmox 2.0, under what circumstances should we use the write-back / write-thru cache?
Should we set to no-cache if using raidcard that already have build in hardware cache?

Thank you.
 
yes, for 2.0 we put that option on the gui. on 1.9 you needed to configure this via CLI, in /etc/qemu-server/VMID.conf.

i prefer cache=no.
 
where are your VM disks? which format?
 
name: testw2k8
ide2: none,media=cdrom
vlan0: rtl8139=AE:87:67:C3:DF:27
ostype: w2k8
bootdisk: virtio0
virtio0: local:103/vm-103-disk-1.raw
memory: 2048
sockets: 1
boot: dc
freeze: 0
cpuunits: 1000
acpi: 1
kvm: 1
onboot: 0
cores: 2
args: -cpu host

cpu: host
net0: virtio=AE:87:67:C3:DF:27,bridge=vmbr0
net1: rtl8139=26:06:CA:6E:B8:CA,bridge=vmbr0
 
I run cache=no on all my boxes for this config.
 
You must log in or register to reply here.
Top Bottom