Content-Type: multipart/mixed; boundary="===============3157973405659946134==" MIME-Version: 1.0 From: Huang Ying To: lkp@lists.01.org Subject: Re: [Btrfs] 3a8b36f3780: -62.6% fileio.requests_per_sec Date: Tue, 31 Mar 2015 16:32:47 +0800 Message-ID: <1427790767.17170.95.camel@intel.com> In-Reply-To: <55094E06.2070004@suse.com> List-Id: --===============3157973405659946134== Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable Hi, Filipe, On Wed, 2015-03-18 at 10:05 +0000, Filipe Manana wrote: [snip] > Hi, thanks for this. > = > However this doesn't make sense to me. > This commit only touches btrfs' fsync handler and the test uses sysbench > without passing --file-fsync-freq to it, which means sysbench will never > do fsyncs according to its man page (default for fsync frequency is 0). > = > Or maybe I missed something? Sorry for late. I checked source code of sysbench and found that the actual default value of --file-fsync-freq is 100 instead of 0 in man page, as in the following lines. {"file-fsync-freq", "do fsync() after this number of requests (0 - don't = use fsync())", SB_ARG_TYPE_INT, "100"}, I double checked that via a debug patch to sysbench too. Best Regards, Huang, Ying --===============3157973405659946134==--