From mboxrd@z Thu Jan 1 00:00:00 1970 From: Bart Van Assche Subject: Re: [PATCH] scsi: default to scsi-mq Date: Tue, 11 Jul 2017 15:46:53 +0000 Message-ID: <1499788012.2586.12.camel@wdc.com> References: <20170616082755.22832-1-hch@lst.de> <1499701840.3555.7.camel@wdc.com> <1499779970.3345.1.camel@wdc.com> <4cf7ed21-7cf4-a91a-8beb-ba5f92e4eaaf@huawei.com> Mime-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable Return-path: Received: from esa1.hgst.iphmx.com ([68.232.141.245]:23754 "EHLO esa1.hgst.iphmx.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S933520AbdGKPq4 (ORCPT ); Tue, 11 Jul 2017 11:46:56 -0400 In-Reply-To: <4cf7ed21-7cf4-a91a-8beb-ba5f92e4eaaf@huawei.com> Content-Language: en-US Content-ID: Sender: linux-scsi-owner@vger.kernel.org List-Id: linux-scsi@vger.kernel.org To: "hch@lst.de" , "linux-scsi@vger.kernel.org" , "john.garry@huawei.com" , "linuxarm@huawei.com" On Tue, 2017-07-11 at 15:14 +0100, John Garry wrote: > On 11/07/2017 14:32, Bart Van Assche wrote: > > On Tue, 2017-07-11 at 11:22 +0100, John Garry wrote: > > > On 10/07/2017 16:50, Bart Van Assche wrote: > > > > Since a fix for the performance regression triggered by this patch = will be upstream > > > > soon (see also https://git.kernel.org/pub/scm/linux/kernel/git/axbo= e/linux-block.git/commit/?h=3Dfor-linus&id=3D32825c45ff8f4cce937ab85b030dc6= 93ceb1aa0a): > > > >=20 > > >=20 > > > FYI, on linux-next 20170711 (which now includes the above patch Bart > > > mentioned) we see a large performance regression on hisi_sas (LLDD do= es > > > not config shost for mq). > > >=20 > > > fio read mode iops goes from ~833K (scsi_mod.use_blk_mq=3Dn on cmdlin= e) to > > > ~320K > >=20 > > Hello John, > >=20 > > Thanks for the feedback. Is the kernel config with which these measurem= ents > > were performed available somewhere? >=20 > It is the default arm64 defconfig with the following changes: > CONFIG_ARM_SMMU_V3=3Dn > CONFIG_9P_FS=3Dn >=20 > We were getting a compile error in the 9p fs code, so disabled it.=20 > Turning on the IOMMU drops performance across the board for our=20 > platform, so just disabling it for the test. Hello John, What block driver controls the block device for which the performance regre= ssion has been observed? How many hardware queues were created by that block driv= er (see also /sys/block/*/mq/...)? I'm asking this because the number of hardw= are queues controls which I/O scheduler is selected as default. From block/elev= ator.c: if (q->mq_ops) { if (q->nr_hw_queues =3D=3D 1) e =3D elevator_get("mq-deadline", false); if (!e) return 0; } else e =3D elevator_get(CONFIG_DEFAULT_IOSCHED, false); Bart.=