From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752319AbbANKR3 (ORCPT ); Wed, 14 Jan 2015 05:17:29 -0500 Received: from youngberry.canonical.com ([91.189.89.112]:40583 "EHLO youngberry.canonical.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751375AbbANKR1 (ORCPT ); Wed, 14 Jan 2015 05:17:27 -0500 MIME-Version: 1.0 In-Reply-To: <20150113162307.GA21338@infradead.org> References: <1421163888-21452-1-git-send-email-ming.lei@canonical.com> <20150113162307.GA21338@infradead.org> Date: Wed, 14 Jan 2015 18:17:24 +0800 Message-ID: Subject: Re: [PATCH v2 0/4] block & aio: improve loop with kernel aio From: Ming Lei To: Christoph Hellwig Cc: Linux Kernel Mailing List , Dave Kleikamp , Jens Axboe , Zach Brown , Maxim Patlasov , Andrew Morton , Alexander Viro , Benjamin LaHaise Content-Type: text/plain; charset=UTF-8 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Wed, Jan 14, 2015 at 12:23 AM, Christoph Hellwig wrote: > On Tue, Jan 13, 2015 at 11:44:44PM +0800, Ming Lei wrote: >> Follows benefits from using kernel aio in loop: >> - avoid double cache, and memory usage decreased a lot >> - system load gets much decreased > > This seems to conflate two changes: > > 1) use direct I/O > 2) use aio (require 1) > > what numbers do you get when just using direct I/O? I don't run the direct I/O only test because it needs quite changes on current patches, such as: make the kiocb as sync, allow all dio requests running concurrently for sake of throughput. IMO, without aio, context switches will be increased inevitably, and CPU will be wasted, so could you explain if there is advantage of not using aio? Thanks, Ming Lei