From mboxrd@z Thu Jan 1 00:00:00 1970 From: ebiederm@xmission.com (Eric W. Biederman) Subject: Re: [PATCH 0/7] Batched netns improvements. Date: Thu, 03 Dec 2009 05:23:05 -0800 Message-ID: References: <1259845591.3766.30.camel@bigi> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: David Miller , netdev@vger.kernel.org, Daniel Lezcano To: hadi@cyberus.ca Return-path: Received: from out02.mta.xmission.com ([166.70.13.232]:42708 "EHLO out02.mta.xmission.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752290AbZLCNXG (ORCPT ); Thu, 3 Dec 2009 08:23:06 -0500 In-Reply-To: <1259845591.3766.30.camel@bigi> (jamal's message of "Thu\, 03 Dec 2009 08\:06\:31 -0500") Sender: netdev-owner@vger.kernel.org List-ID: jamal writes: > On Thu, 2009-12-03 at 04:27 -0800, Eric W. Biederman wrote: >> This series of patches takes my simple 4k network namespace exit test >> from about 44s to 1.6s, with a minuscule increase in code side. > > Nice. > I will test these as soon as they make it into net-next (just easier > for me given the setup). > I should confirm that: > The previous patches which are already in net-next from yesterday do > improve performance to under 1 minute. Can we call that a several-factor > improvement?;-> > > Are you planning to give some tender loving to the virtual devices > as well (gre etc) or is that taken care of now you have a pernet batch > exit op? The tunnel devices gre, ipip, sit, ip6_tunnel, and the ethernet bridging still need some tlc to give them rtnl_link_ops, and in particular dellink operations so the deletions can be done in one big batch. The batch operations are mostly a generalization of a bunch of what was already there. I keep planning on doing something else and then I wind up working on this. Mostly what I have done turned out to be some very low hanging fruit, and some general cleanups. Eric