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

[David Miller <davem@davemloft.net>]

From: Peter Zijlstra <a.p.zijlstra@chello.nl>
Date: Tue, 07 Oct 2008 10:31:30 +0200

> On Mon, 2008-10-06 at 14:40 -0700, David Miller wrote:
> > From: Eric Dumazet <dada1@cosmosbay.com>
> > 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% ?

Probably the RCU delay on a 128 cpu machine :-)

Also I bet batching the socket destruction eliminates all of
the cached local state we have in the cpu at the actual socket
destruction time.