From mboxrd@z Thu Jan 1 00:00:00 1970 From: Andi Kleen Subject: Re: [RFC] Use RCU for tcp_ehash lookup Date: Thu, 2 Sep 2004 23:19:50 +0200 Sender: netdev-bounce@oss.sgi.com Message-ID: <20040902211950.GH16175@wotan.suse.de> References: <20040831125941.GA5534@in.ibm.com> <20040831135419.GA17642@wotan.suse.de> <20040901113641.GA3918@in.ibm.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Andi Kleen , davem@redhat.com, netdev@oss.sgi.com, linux-kernel@vger.kernel.org, Dipankar , paulmck@us.ibm.com Return-path: To: Srivatsa Vaddagiri Content-Disposition: inline In-Reply-To: <20040901113641.GA3918@in.ibm.com> Errors-to: netdev-bounce@oss.sgi.com List-Id: netdev.vger.kernel.org On Wed, Sep 01, 2004 at 05:06:41PM +0530, Srivatsa Vaddagiri wrote: > | 2.6.8.1 | 2.6.8.1 + my patch > ------------------------------------------------------------------------------- > Average cycles | | > spent in | | > __tcp_v4_lookup_established | 2970.65 | 668.227 > | (~3.3 micro-seconds) | (~0.74 microseconds) > ------------------------------------------------------------------------------- > > This repesents improvement by a factor of 77.5%! Nice. > > > > > > And it should also fix the performance problems with > > cat /proc/net/tcp on ppc64/ia64 for large hash tables because the rw locks > > are gone. > > But spinlocks are in! Would that still improve the performance compared to rw > locks? (See me earlier note where I have explained that lookup done for > /proc/net/tcp is _not_ lock-free yet). Yes, spinlocks are much faster than rwlocks. -Andi