From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S266717AbUH2Agi (ORCPT ); Sat, 28 Aug 2004 20:36:38 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S267457AbUH2Agf (ORCPT ); Sat, 28 Aug 2004 20:36:35 -0400 Received: from hera.cwi.nl ([192.16.191.8]:16324 "EHLO hera.cwi.nl") by vger.kernel.org with ESMTP id S266717AbUH2Age (ORCPT ); Sat, 28 Aug 2004 20:36:34 -0400 Date: Sun, 29 Aug 2004 02:36:32 +0200 From: Andries Brouwer To: linux-kernel@vger.kernel.org Subject: orlov/nobh Message-ID: <20040829003630.GA19197@apps.cwi.nl> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.4i Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org > > +oldalloc Use old allocator for new inodes. > > +orlov (*) Use Orlov allocator. > > + > > +nobh Do not attach buffer_heads to file pagecache. > > For these two it would be nice to include a description > of why you'd want them or a pointer to something > describing it. An admin trying to squeeze performance out > of his server might see these options in such documentation > and then want to know if they can help him. See http://lwn.net/Articles/14633/ and http://lwn.net/Articles/14446/ for Orlov. I am not aware of serious benchmark work. Simple direct tests seem either to show no significant difference or to favor Orlov. I have never seen a study on fragmentation after extended use of both allocators. Concerning nobh, I think that is a toy by Andrew, meant for very large machines. He is in a better position to comment (in fact on both orlov and nobh). Andries