new fedora virtio-win-0.1-22.iso drive (february 2012)

[spirit]

http://alt.fedoraproject.org/pub/alt/virtio-win/latest/images/bin/

please test !

 

[tom]

I run all my boxes with these drivers since 2 days, no issues so far.

 

[e100]

I am seeing significant increases in performance with no issues so far.

 

[spirit]

Same here !
writes are faster and before I had BSOD with cache=none, now it's resolved

 

[e100]

Here is a benchmark I posted last week in another thread using virtio drivers:

Upgraded to 0.1-22 today and ran a new benchmark on that exact same VM.
Both benchmarks were done with cache=none LVM on top of DRBD over Infiniband Areca 1880ix-12 4GB BBU with 6 500GB Segate SATA III disk in RAID 5.


Still not native speed but much closer!

 

[spirit]

nice !

If you have a fast storage, you can also try to pass "ioeventfd=off" to your drive option.
sometimes it help. (you need low latency storage)

 

[e100]

nice !

If you have a fast storage, you can also try to pass "ioeventfd=off" to your drive option.
sometimes it help. (you need low latency storage)

I could not figure out how to add that to the config file, but I manually started the VM and was able to add it.

Sequential write is way better, the rest is mixed results.

Before ioeventfd=off:



After ioeventfd=off:


My sequential writes natively are limited to about 700MB/sec once I exceed the cache ram on my RAID card which this benchmark did.

 

[syd82]

http://alt.fedoraproject.org/pub/alt/virtio-win/latest/images/bin/

please test !

Viostor crashes about once a day, without an apparent reason.

OS: W2k3 SP2 32 Bit
Driver Version: 52.63.103.2200

Here's the relevant information:

==================================================
Dump File : Mini030312-06.dmp
Crash Time : 03/03/2012 10.14.01
Bug Check String : DRIVER_IRQL_NOT_LESS_OR_EQUAL
Bug Check Code : 0x000000d1
Parameter 1 : 0x00000160
Parameter 2 : 0xd0000002
Parameter 3 : 0x00000000
Parameter 4 : 0xf773078e
Caused By Driver : viostor.sys
Caused By Address : viostor.sys+178e
File Description : Red Hat VirtIO SCSI driver
Product Name : Red Hat VirtIO SCSI controller
Company : Red Hat Inc.
File Version : 52.63.103.2200 built by: WinDDK
Processor : 32-bit
Crash Address : ntkrnlpa.exe+8c9eb
Stack Address 1 : viostor.sys+178e
Stack Address 2 : ntkrnlpa.exe+3211c
Stack Address 3 : ntkrnlpa.exe+8dba7
Computer Name :
Full Path : C:\WINDOWS\Minidump\Mini030312-06.dmp
Processors Count : 4
Major Version : 15
Minor Version : 3790
Dump File Size : 67.880
==================================================


Any thoughts ?

BR,
Simone.

 

[e100]

What cache setting are you using for your disks?
I have had issues when using writeback and always use cache=none.

That error can also be caused by bad RAM.
It could be a driver issue too, but I am not running any virtio drivers on Win 32bit so I have no idea if it is stable or not.

 

[syd82]

The disks are all LVM-backed with cache=none. Hardware is not the issue, because it's a cluster comprised of two nodes and I tried running the VM on both.

Regards,
Simone.

 

[spirit]

you could also try passing "elevator=deadline" to your grub

this is the best scheduler for virtio

edit /etc/default/grub

GRUB_CMDLINE_LINUX_DEFAULT="elevator=deadline"

#update-grub
reboot

 

[syd82]

you could also try passing "elevator=deadline" to your grub

this is the best scheduler for virtio

edit /etc/default/grub

GRUB_CMDLINE_LINUX_DEFAULT="elevator=deadline"

#update-grub
reboot

I will certainly do, but I'm missing something...how does I/O scheduler affects guest stability ?

Regards,
Simone.