From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755949AbdKJE1R (ORCPT ); Thu, 9 Nov 2017 23:27:17 -0500 Received: from mail.kernel.org ([198.145.29.99]:56472 "EHLO mail.kernel.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755926AbdKJE1Q (ORCPT ); Thu, 9 Nov 2017 23:27:16 -0500 DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org 7325A21879 Authentication-Results: mail.kernel.org; dmarc=none (p=none dis=none) header.from=kernel.org Authentication-Results: mail.kernel.org; spf=none smtp.mailfrom=shli@kernel.org Date: Thu, 9 Nov 2017 20:27:13 -0800 From: Shaohua Li To: Tejun Heo Cc: Jens Axboe , linux-kernel@vger.kernel.org, kernel-team@fb.com Subject: Re: [PATCH 1/2] blk-throtl: make latency= absolute Message-ID: <20171110042713.quytcwbu6g6xwvpt@kernel.org> References: <20171109221924.GB983427@devbig577.frc2.facebook.com> <20171109231212.mbqwyzpmciyshxov@kernel.org> <20171109234258.GD983427@devbig577.frc2.facebook.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20171109234258.GD983427@devbig577.frc2.facebook.com> User-Agent: NeoMutt/20170609 (1.8.3) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Thu, Nov 09, 2017 at 03:42:58PM -0800, Tejun Heo wrote: > Hello, Shaohua. > > On Thu, Nov 09, 2017 at 03:12:12PM -0800, Shaohua Li wrote: > > The percentage latency makes sense, but the absolute latency doesn't to me. A > > 4k IO latency could be much smaller than 1M IO latency. If we don't add > > baseline latency, we can't specify a latency target which works for both 4k and > > 1M IO. > > It isn't adaptive for sure. I think it's still useful for the > following reasons. > > 1. The absolute latency target is by nature both workload and device > dependent. For a lot of use cases, coming up with a decent number > should be possible. > > 2. There are many use cases which aren't sensitive to the level where > they care much about the different between small and large > requests. e.g. protecting a managerial job so that it doesn't > completely stall doesn't require tuning things to that level. A > value which is comfortably higher than usually expected latencies > would often be enough (say 100ms). > > 3. It's also useful for verification / testing. I think the absolute latency would only work for HD. For a SSD, a 4k latency probably is 60us and 1M latency is 500us. The disk must be very contended to make 4k latency reach 500us. Not sensitive doesn't mean no protection. If the use case sets rough latency, say 1ms, there will be no protection for 4k IO at all. The baseline latency is pretty reliable for SSD actually. So I'd rather keeping the baseline latency for SSD but using absolute latency for HD, which can be done easily by setting DFL_HD_BASELINE_LATENCY to 0. Thanks, Shaohua