From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1030421Ab2JSPsO (ORCPT ); Fri, 19 Oct 2012 11:48:14 -0400 Received: from mx1.redhat.com ([209.132.183.28]:46650 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1030220Ab2JSPsN (ORCPT ); Fri, 19 Oct 2012 11:48:13 -0400 Date: Fri, 19 Oct 2012 11:00:11 -0400 From: Vivek Goyal To: Tejun Heo Cc: Robin Dong , linux-kernel@vger.kernel.org, Robin Dong , Jens Axboe , Tao Ma Subject: Re: [PATCH V3] block/throttle: Add IO throttled information in blkio.throttle Message-ID: <20121019150011.GE27052@redhat.com> References: <1349765625-4020-1-git-send-email-robin.k.dong@gmail.com> <20121016232706.GI16166@google.com> <20121017134945.GC31663@redhat.com> <20121018232404.GW13370@google.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20121018232404.GW13370@google.com> 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 On Thu, Oct 18, 2012 at 04:24:04PM -0700, Tejun Heo wrote: > Hello, Vivek. > > On Wed, Oct 17, 2012 at 09:49:45AM -0400, Vivek Goyal wrote: > > Can you explain a bit more. Whe do you mean by "total number queued". I > > think throttle.io_queued will total number of bios queued in the cgroup > > at the time of query. > > Instead of exposing the number that blk-throttle currently has > deferred, we can expose the number of bios that have been sent to > blk-throttle and the number of bios which left blk-throttle, both > monotically increasing and the difference indicating the number being > deferred. Ok, I see it now. So we currently already maintain the number of IOs dispatched from blk-throttle in throttle.io_serviced. Now you are suggesting that maintain another counter which keeps track of total number IOs submitted to blk-throttle, say throttle.io_submitted? I think using throttle.io_queued will be little confusing because in CFQ we already use blkio.io_queued to represent number of IOs currently queued and it is not monotonically increasing value. > That way we can stick to the usual stats facility. So how does this help? Because it is a monotonically increasing value we can use per cpu stats without extra locking? Or somthing else? Thanks Vivek