Xen devel - May 2007 - [PATCH] Fix the very-slow-IDE-IO-speed issue of Qemu 0.90

c/s 15021 updated Qemu to 0.90. In the new Qemu''s ide.c, asynchronous
IO
(AIO) is used to replace the previous DMA thread; here when an AIO
request is completed, dom0 sends a signal SIGUSR2 to Qemu (see
block-raw.c: qemu_aio_init(),raw_aio_setup()), then the signal
interrupts the select() in main_loop_wait() at once, next,
qemu_aio_poll() is called to reap a completed AIO request.

However, in certain Linux distributions (i.e., x86_64 RHEL 4u4), if Qemu
is spawned by the python script image.py of Control Panel, the SIGUSR2
of Qemu is blocked by default due to some reasons (i.e. it seems Python
2.3.4 has this issue; 2.4.2 is ok); so in most cases the select() in
main_loop_wait() can only time out in 10ms, then the qemu_aio_poll() is
called - this results in a very slow disk IO speed...

The attached patch ensures aio_sig_num is unblocked when the AIO is used
in Qemu 0.90.

Signed-off-by: Dexuan Cui <dexuan.cui@intel.com>



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

> c/s 15021 updated Qemu to 0.90. In the new Qemu''s ide.c,
asynchronous
IO
> (AIO) is used to replace the previous DMA thread; here when an AIO
> request is completed, dom0 sends a signal SIGUSR2 to Qemu (see
> block-raw.c: qemu_aio_init(),raw_aio_setup()), then the signal
> interrupts the select() in main_loop_wait() at once, next,
> qemu_aio_poll() is called to reap a completed AIO request.
> 
> However, in certain Linux distributions (i.e., x86_64 RHEL 4u4), if
> Qemu is spawned by the python script image.py of Control Panel, the
SIGUSR2
> of Qemu is blocked by default due to some reasons 
Good work figuring this out! It perfectly explains why most folk
weren''t
seeing the problem.

Have you got any comparative performance numbers comparing IDE
performance before and after the 0.90 upgrade (with your fix)? Are we
back to roughly where we were?

Thanks,
Ian   


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

I did some simple disk IO tests. Using c/s 15021 plus my fix, I''m sure
the disk performance does back to roughly where we were.

-- Dexuan

-----Original Message-----
From: Ian Pratt [mailto:Ian.Pratt@cl.cam.ac.uk] 
Sent: 2007年5月29日 13:26
To: Cui, Dexuan; xen-devel
Cc: ian.pratt@cl.cam.ac.uk
Subject: RE: [Xen-devel] [PATCH] Fix the very-slow-IDE-IO-speed issue of
Qemu0.90
> Have you got any comparative performance numbers comparing IDE
> performance before and after the 0.90 upgrade (with your fix)? Are we
> back to roughly where we were?
Ian   

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel