Re: [PATCH 3/3] Convert the UDP hash lock to RCU

[Paul E. McKenney]

On Tue, Oct 07, 2008 at 10:31:30AM +0200, Peter Zijlstra wrote:
> On Mon, 2008-10-06 at 14:40 -0700, David Miller wrote:
> > From: Eric Dumazet <dada1@xxxxxxxxxxxxx>
> > Date: Mon, 06 Oct 2008 23:22:31 +0200
> > 
> > > Me wondering what impact this synchronize_rcu() can have on mono-threaded
> > > VOIP applications using lot of UDP sockets. What is the maximum delay of
> > > this function ?
> > 
> > The cost is enormous, we really can't use it here.
> > 
> > I have a patch that did top-level socket destruction using RCU,
> > and that didn't use synchronize_rcu(), and that killed connection
> > rates by up to %20.
> 
> Did you ever figure out why you lost those 20% ?
> 
> > I can only imagine what the cost would be if I had to add such a call
> > in there.
> 
> Yeah, sync_rcu() is rediculously expensive, at best 3 jiffies IIRC.

I could make it -much- faster, but at the expense of -serious- CPU
overhead.  Still, might be useful during boot time (when the system
can't do anything useful anyway) to accelerate getting data structures
initialized.

							Thanx, Paul
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/