Re: [Qemu-devel] Re: [PATCH 2/3] Move aio implementation out of raw block driver

[Anthony Liguori <anthony@codemonkey.ws>]

Gerd Hoffmann wrote:
> Anthony Liguori wrote:
>   
>> Gerd Hoffmann wrote:
>>     
>>> How about providing a aio interface implementation which simply uses
>>> read/write syscalls (thereby not being really async obviously)?  Then
>>> use that as fallback instead of aio emulation?  And also drop CONFIG_AIO
>>> then?
>>>       
>> Yeah, this is basically what block-raw-posix does today.  I was thinking
>> the same thing.  I was also thinking that you could do an aio
>> implementation for win32 and possibly reunify block-raw-posix and
>> block-raw-linux.
>>     
>
> Sure, that the next logical steps.  Later we can also convert all
> block-* drivers to the new aio interface and subsequently drop alot of
> dead block layer code.
>   

Yup.

>> But before going down this route, I want to see if linux-aio is really
>> the right tool for the job.
>>     
>
> IMHO this all makes sense even in case linux-aio turns out to not be
> worth it.
>   

Yes, I agree.  For a bit of a spoiler, initial results are that 
cache=off + my fd_pool patch is getting equivalent performance to 
linux-aio so it's looking like we can avoid linux-aio for now.

Regards,

Anthony Liguori

> cheers,
>   Gerd
>
> --
> To unsubscribe from this list: send the line "unsubscribe kvm" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>