Re: [Qemu-devel] [PATCH 0/2] Enable qemu block layer to not flush

[Anthony Liguori <anthony@codemonkey.ws>]

On 05/10/2010 05:03 PM, Alexander Graf wrote:
> On 10.05.2010, at 23:59, Anthony Liguori wrote:
>
>    
>> On 05/10/2010 04:51 PM, Alexander Graf wrote:
>>      
>>> Thanks to recent improvements, qemu flushes guest data to disk when the guest
>>> tells us to do so.
>>>
>>> This is great if we care about data consistency on host disk failures. In cases
>>> where we don't it just creates additional overhead for no net win. One such use
>>> case is the building of appliances in SUSE Studio. We write the resulting images
>>> out of the build VM, but compress it directly afterwards. So if possible we'd
>>> love to keep it in RAM.
>>>
>>> This patchset introduces a new block parameter to -drive called "flush" which
>>> allows a user to disable flushing in odd scenarios like the above. To show the
>>> difference in performance this makes, I have put together a small test case.
>>> Inside the initrd, I call the following piece of code on a 500MB preallocated
>>> vmdk image:
>>>
>>>        
>> This seems like it's asking for trouble to me.  I'm not sure it's worth the minor performance gain.
>>      
> The gain is little on my netbook where I did the test on. This is part of performance regressions from 0.10 to 0.12 where we're talking build times of 2 minutes going to 30. While writeback was most of the chunk, flushing still at least doubled the build times which is unacceptable for us.
>    

There's got to be a better place to fix this.  Disable barriers in your 
guests?

Regards,

Anthony Liguori

> I also fail to see where it's asking for trouble. If we don't flush volatile data, things are good, no?
>
>
> Alex
>
>