Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753242AbdHBM3F (ORCPT ); Wed, 2 Aug 2017 08:29:05 -0400 Received: from mx2.suse.de ([195.135.220.15]:52736 "EHLO mx1.suse.de" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1752436AbdHBM3E (ORCPT ); Wed, 2 Aug 2017 08:29:04 -0400 Date: Wed, 2 Aug 2017 14:29:01 +0200 From: Michal Hocko To: Philipp Hahn Cc: linux-mm@kvack.org, "linux-kernel@vger.kernel.org" Subject: Re: [BUG] Slow SATA disk - waiting in balance_dirty_pages() on i686-pae.html Message-ID: <20170802122901.GB2529@dhcp22.suse.cz> References: <2a526c85-0e27-7a5d-f606-66c74499352a@univention.de> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <2a526c85-0e27-7a5d-f606-66c74499352a@univention.de> User-Agent: Mutt/1.5.23 (2014-03-12) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 677 Lines: 19 On Tue 01-08-17 12:32:40, Philipp Hahn wrote: > Hello, > > ;TL,DR: apt-get is blocked by balance_dirty_pages() with linux-{3,16 > 4.2, 4.9}, but fast after reboot. > > > We still have several systems running 4.9.0-ucs104-686-pae. They have 16 > GiB RAM and two disk: I would strongly discourage you from using 32b system with so much memory. This will always bump into problems because of the inherent kernel/userspace split. Also I would bet that the problem you are seeing is the lack of lowmem memory which is considered for the dirty writers throttling unless you have vm.highmem_is_dirtyable is set to 1 which is not the case in your setup. -- Michal Hocko SUSE Labs