On 08/01/2011 04:17 PM, Avi Kivity wrote:
On 07/29/2011 06:25 PM, Sasha Levin wrote:
On Fri, 2011-07-29 at 20:01 +0800, Liu Yuan wrote:
> Looking at this long list,most are function pointers that can not be
> inlined, and the internal data structures used by these functions are
> dozons. Leave aside code complexity, this long code path would really
> need retrofit. As Christoph simply put, this kind of mess is inherent
> all over the qemu code. So I am afraid, the 'retrofit' would end
up to
> be a re-write the entire (sub)system. I have to admit that, I am
> inclined to the MST's vhost approach, that write a new subsystem
other
> than tedious profiling and fixing, that would possibly goes as far as
> actually re-writing it.
I don't think the fix for problematic userspace is to write more kernel
code.
vhost-net improved throughput and latency by several factors, allowing
to achieve much more than was possible at userspace alone.
With vhost-blk we see an improvement of ~15% - which I assume by your
and Christoph's comments can be mostly attributed to QEMU. Merging a
module which won't improve performance dramatically compared to what is
possible to achieve in userspace (even if it would require a code
rewrite) sounds a bit wrong to me
Agree. vhost-net works around the lack of async zero copy networking
interface. Block I/O on the other hand does have such an interface,
and in addition transaction rates are usually lower. All we're saving
is the syscall overhead.
Personally I too agree with Sasha Levin. But vhost-blk is the first fast
prototype that is supposed to act as a code base to do further
optimisation, which I plan to utilize kernel's internal stuff like BIO
layer, that can not be accessed from user space, to maximize the
performance for raw disk based block IO.
Yuan
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html