From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932961AbaGUPzn (ORCPT ); Mon, 21 Jul 2014 11:55:43 -0400 Received: from 8bytes.org ([81.169.241.247]:51566 "EHLO theia.8bytes.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932420AbaGUPzm (ORCPT ); Mon, 21 Jul 2014 11:55:42 -0400 Date: Mon, 21 Jul 2014 17:55:40 +0200 From: Joerg Roedel To: Pavel Machek Cc: "Rafael J. Wysocki" , Len Brown , linux-pm@vger.kernel.org, linux-kernel@vger.kernel.org Subject: Re: [PATCH 0/6] PM / Hibernate: Memory bitmap scalability improvements Message-ID: <20140721155540.GO30979@8bytes.org> References: <1405684643-30234-1-git-send-email-joro@8bytes.org> <20140719134904.GA2852@amd.pavel.ucw.cz> <20140721083211.GK30979@8bytes.org> <20140721114507.GA13921@amd.pavel.ucw.cz> <20140721142951.GN30979@8bytes.org> <20140721143905.GA16454@amd.pavel.ucw.cz> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20140721143905.GA16454@amd.pavel.ucw.cz> User-Agent: Mutt/1.5.21 (2010-09-15) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hi, On Mon, Jul 21, 2014 at 04:39:05PM +0200, Pavel Machek wrote: > So you reduced O(n) to O(n/2 log(n)) ... that's still O(n log(n)), right? Radix trees have a memory complexity of ~O(n + log(n)). We have an additional log(n) space requirement for the tree itself, but we linearly safe memory for each node compared to the old implementation. Since n beats log(n) at some point there is a point where we need less pages with the new code than with the old. That break-even point is somewhere around 32GB of RAM, for smaller sizes we need 2 additional pages at most. > > > You can also modify the above tool to give you some data to plot, then > > you will also SEE that there is still a linear relationship between RAM > > size and memory required by the old and new memory bitmap > > implementation. > > I believe radix tree is O(log(N)) per operation, but its memory is > certainly not linear. Numbers beat believes. Joerg