Return-Path: Received: from aserp1040.oracle.com ([141.146.126.69]:35203 "EHLO aserp1040.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752803AbcCGPwU (ORCPT ); Mon, 7 Mar 2016 10:52:20 -0500 To: Andreas Dilger Cc: Martin Petersen , Jens Axboe , linux-fsdevel , linux-block@vger.kernel.org, calvinowens@fb.com, "Theodore Ts'o" , Christoph Hellwig Subject: Re: [PATCH 0/11] Update version of write stream ID patchset From: "Martin K. Petersen" References: <1457107853-8689-1-git-send-email-axboe@fb.com> <56AF7A2D-DF4F-4973-B4F6-6F515A8C500C@dilger.ca> <59389FFA-5207-4B21-9614-F080CD24DC65@dilger.ca> Date: Mon, 07 Mar 2016 10:52:05 -0500 In-Reply-To: <59389FFA-5207-4B21-9614-F080CD24DC65@dilger.ca> (Andreas Dilger's message of "Sun, 6 Mar 2016 15:42:27 -0700") Message-ID: MIME-Version: 1.0 Content-Type: text/plain Sender: linux-fsdevel-owner@vger.kernel.org List-ID: >>>>> "Andreas" == Andreas Dilger writes: Andreas> It doesn't cost anything to reserve the first 32 values for Andreas> filesystem metadata, and they can be aggregated more or less Andreas> depending on the hardware capabilities. Even if current Andreas> devices only support 4 or 8 streams, I'm sure that this will Andreas> improve in the future, so it doesn't make sense to limit Andreas> ourselves based on the very first devices on the market. I am not against reserving parts of the namespace. As long as the anticipation is that those 32 values in practice will be consolidated into probably 2 (journal and metadata) in a static fashion. Because, as I said, there are performance and serialization issues wrt. closing and opening streams. So I don't think doing it in the hotpath is going to fly. Andreas> Also, this opens up interesting possibilities for blktrace, DM Andreas> layers like dm-thinp, bcache, etc. that are currently lacking Andreas> any kind of data on how they should allocate blocks. Ted Andreas> described the contortions he does to map from block offsets in Andreas> blktrace to filesystem metadata using debugfs output and Andreas> scripts, and not everyone is as knowledgeable about filesystem Andreas> internals as he is, but still wants to be able to diagnose Andreas> filesystem IO latency issues. I am a big proponent of I/O tagging. We use it extensively in various products and it works beautifully. That's why I would like to see the storage standards being able to support it. -- Martin K. Petersen Oracle Linux Engineering