Re: [Qemu-devel] [RFC] Replace posix-aio with custom thread pool

[Anthony Liguori <anthony@codemonkey.ws>]

Gerd Hoffmann wrote:
> Andrea Arcangeli wrote:
>   
>> My current feeling is that this user thread aio thing will never
>> satisfy enterprise usage and kernel aio is mandatory in my view.
>>     
>
> Well, linux kernel aio has its share of problems too:
>
>   * Anthony mentioned it may block on certain circumstances (forgot
>     which ones), and you can't figure beforehand to turn off aio then.
>
>   * It can't handle block allocation.  Kernel handles that by doing
>     such writes synchronously via VFS layer (instead of the separate
>     aio code paths).  Leads to horrible performance and bug reports
>     such as "installs on sparse files are very slow".
>
>   * support for vectored aio isn't that old.  IIRC it was added
>     somewhen around 2.6.20 (newer that current suse/redhat enterprise
>     versions).  Which IMHO means you can't expect it being present
>     unconditionally.
>
>   
>> And we should concentrate on kernel aio and get rid
>> of threads when host OS is linux.
>>     
>
> Threads will be there anyway for kvm smp.
>
>   
>> Has anybody a patch implementing kernel aio that I can plug into the
>> dma zerocopy api? I'm not so sure clone aio is worth maintaining
>> inside qemu instead of evolving glibc
>>     
>
> Well, wait for glibc isn't going to fly.  glibc waits for posix, and
> posix waits for a reference implementation (which will not be glibc).
>
>   
>> and kernel with preadv/pwritev
>>     
>
> With that in place you don't need kernel aio any more, then you can
> really do it in userspace with threads.  But that probably would be
> linux-only  ^W^W^W
>   

linux-only is okay but we just need a relatively sane fall back.  There 
have been preadv/pwritev patches posted before, they just for some 
reason never were merged.

http://lwn.net/Articles/163603/

> ahem: http://www.daemon-systems.org/man/preadv.2.html
>   

Yeah, dunno if that's all BSDs or just NetBSD.

Regards,

Anthony Liguori

> cheers,
>   Gerd
>