Hi, In our application we are seeing occasional high latency spikes (500ms range )when writing small amounts of data (a couple MB, a couple of 100 IOs) to files on an EXT4 volume (data=ordered). We are running kernel 3.10.62. What seems to be happening is that a journal commit takes a long time. During that time no write access to the journal is possible. So a subsequent vfs.write is blocked until the journal commit is finished. We expected the page cache to decouple the vfs.write from this and not to block the vfs.write. This behavior causes large spikes in on average much lower write latencies. Small log excerpt (duration in us): 16:18:56.966387, PROCA, 20234, __ext4_journal_get_write_access: duration: 140412 16:18:56.966406, PROCB, 19965, __ext4_journal_get_write_access: duration: 461892 16:18:56.966422, PROCC, 20387, __ext4_journal_get_write_access: duration: 391105 16:18:56.966439, PROCD, 20041, __ext4_journal_get_write_access: duration: 64256 16:18:56.966454, itheex, 19823, __ext4_journal_get_write_access: duration: 383671 16:18:56.966718, jbd2/dm-8-8, 684, jbd2_journal_commit_transaction: duration: 468959 16:18:56.967426, PROCE, 19867, ext4_file_write: duration: 469431 16:18:56.967442, PROCE, 19867, vfs.write: duration: 469452 A similar, but less frequent issue is with the __ext4_journal_start_sb. Small log excerpt: 16:17:43.337004, PROCF, 20041, __ext4_journal_start_sb: duration: 240753 16:17:43.337024, PROCG, 19980, __ext4_journal_start_sb: duration: 269051 16:17:43.337498, PROCG, 19980, ext4_file_write: duration: 269531 16:17:43.337515, PROCG, 19980, vfs.write: duration: 269554 Can anything be done about these very high latency spikes ? Best regards, Jean-Paul Smeets -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
