From: Shehjar Tikoo Subject: ext4 pre-allocation performance through nfsd Date: Sat, 19 Jul 2008 10:37:16 +1000 Message-ID: <4881373C.10404@cse.unsw.edu.au> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit To: linux-ext4@vger.kernel.org Return-path: Received: from tone.orchestra.cse.unsw.EDU.AU ([129.94.242.59]:56332 "EHLO tone.orchestra.cse.unsw.EDU.AU" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751742AbYGSAvu (ORCPT ); Fri, 18 Jul 2008 20:51:50 -0400 Received: From [203.143.160.170] ([203.143.160.170] == chaos.ken.nicta.com.au) (auth-user shehjart) (cse-authentic-sender shehjart) (for ) By tone With Smtp ; Sat, 19 Jul 2008 10:51:47 +1000 Sender: linux-ext4-owner@vger.kernel.org List-ID: Hi All FYI, I ran some performance tests for a feature that incorporates pre-allocation support into linux nfsd. The patch is a proof-of-concept but the measurements are still relevant for this list. 1. Sequential read throughput with iozone reading a file written using various pre-allocation sizes through nfsd: http://www.gelato.unsw.edu.au/~shehjart/docs/nfsmeasurements/ext4fallocate_read.png 2. Sequential write throughput with iozone over NFS mount with the nfsd writing the file contents to disk using pre-allocation. http://www.gelato.unsw.edu.au/~shehjart/docs/nfsmeasurements/ext4fallocate_write.png Please see the following thread for more details. http://thread.gmane.org/gmane.linux.nfs/21365 From the write throughput plot, its clear that pre-allocation would be highly beneficial if the pre-allocation overhead can be reduced. Thanks Shehjar