From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S263262AbTJQAVI (ORCPT ); Thu, 16 Oct 2003 20:21:08 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S263263AbTJQAVI (ORCPT ); Thu, 16 Oct 2003 20:21:08 -0400 Received: from bi01p1.co.us.ibm.com ([32.97.110.142]:37546 "EHLO DYN320019.beaverton.ibm.com") by vger.kernel.org with ESMTP id S263262AbTJQAVG (ORCPT ); Thu, 16 Oct 2003 20:21:06 -0400 Date: Thu, 16 Oct 2003 10:17:11 -0700 From: "Paul E. McKenney" To: "M. Fioretti" Cc: linux-kernel@vger.kernel.org Subject: Re: Kernel 2.6: new features and older computers Message-ID: <20031016171711.GA1676@us.ibm.com> Reply-To: paulmck@us.ibm.com References: <20031006052154.GB26388@inwind.it> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20031006052154.GB26388@inwind.it> User-Agent: Mutt/1.4.1i Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Hello, Marco, I can help with one of these... On Mon, Oct 06, 2003 at 07:21:54AM +0200, M. Fioretti wrote: > 4) What is RCU and which kind of performances it will improve RCU is a locking mechanism that is useful primarily for read-mostly data structures. It can be thought of as a reader-writer lock in which the readers need not actually do anything, which means that you can get substantial performance improvements in read-mostly situations. The writers must perform updates in a careful manner to avoid messing up the readers. In particular, if a writer removes an element from a list, it must wait for a "grace period" before freeing up the element, since readers might still be referencing it. RCU provides primitives that wait for grace periods to elapse. The best introduction to RCU is probably: http://www.linuxjournal.com/article.php?sid=6993 More information on RCU, including performance comparisons in the Linux kernel may be found at: http://www.rdrop.com/users/paulmck/rclock/ Thanx, Paul