From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-qa0-f54.google.com (mail-qa0-f54.google.com [209.85.216.54]) (using TLSv1 with cipher RC4-SHA (128/128 bits)) (No client certificate requested) by mail09.linbit.com (LINBIT Mail Daemon) with ESMTPS id E34ED1028A4A for ; Thu, 20 Nov 2014 16:38:47 +0100 (CET) Received: by mail-qa0-f54.google.com with SMTP id i13so2041999qae.41 for ; Thu, 20 Nov 2014 07:38:44 -0800 (PST) Sender: Tejun Heo Date: Thu, 20 Nov 2014 10:38:41 -0500 From: Tejun Heo To: Jan Kara Message-ID: <20141120153841.GF14877@htj.dyndns.org> References: <1416299848-22112-1-git-send-email-tj@kernel.org> <1416299848-22112-2-git-send-email-tj@kernel.org> <20141120152702.GF2330@quack.suse.cz> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20141120152702.GF2330@quack.suse.cz> Cc: axboe@kernel.dk, Mike Snitzer , Neil Brown , linux-kernel@vger.kernel.org, Wu Fengguang , Alasdair Kergon , drbd-dev@lists.linbit.com Subject: Re: [Drbd-dev] [PATCH 01/10] writeback: move backing_dev_info->state into bdi_writeback List-Id: "*Coordination* of development, patches, contributions -- *Questions* \(even to developers\) go to drbd-user, please." List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Hello, Jan. On Thu, Nov 20, 2014 at 04:27:02PM +0100, Jan Kara wrote: > Hum, does it make sense to convert BDI_sync_congested and > BDI_async_congested? It contains information whether the *device* is > congested and cannot take more work. I understand that in a cgroup world Yeah, I mean, with cgroup writeback, the device itself doesn't matter. The only thing writeback sees is that cgroup's slice of the device whose congestion status can be independent from other slices of the device. > you want to throttle IO from a cgroup to a device so when you take > bdi_writeback to be a per-cgroup structure you want some indication there > that a particular cgroup cannot push more to the device. But is it that > e.g. mdraid cares about a cgroup and not about the device? I didn't update mdraid to support cgroup writeback yet but it depends on how it's implemented. If it just transmits back the pressure from individual underlying cgroup split devices, it's the same. If we wanna put blkcg splitting in front of mdraid and keep the backend side clear of cgroup splitting, it'd just send down everything as belonging to the root cgroup. Thanks. -- tejun