[sheepdog] [RFC PATCH] sdnet: threading tx/rx process

[Christoph Hellwig]

On Fri, Jul 13, 2012 at 08:25:37PM +0800, Liu Yuan wrote:
> On 07/13/2012 04:47 PM, Christoph Hellwig wrote:
> > On Wed, Jul 11, 2012 at 06:11:31PM +0800, Liu Yuan wrote:
> >> This patch also remove request throttling because we are guaranteed to process
> >> requests as soon as it arrives and won't block any request.
> > 
> > I don't even understand the reason for blocking_conn_list, even with
> > epoll + O_NONBLOCK I/O we should not have to have an artifical limit
> > like that.
> > 
> 
> With short thread, this isn't needed at all, cause requests are never
> blocked.

What did these requests ever get blocked on?  The only reason why
a request goes on blocking_conn_list is because the (arbitrary) maximum
data in flight is exceeded.

> > How do the numbers look if you compare just removing
> > MAX_OUTSTANDING_DATA_SIZE to your full patch?
> > 
> 
> I think this macro matters nothing to the performance, it should be
> never reached for now.

256MB it not a lot for a gateway that has a couple of VMs connected.