f2fs: fix to write node pages with WRITE_SYNC
authorJaegeuk Kim <jaegeuk.kim@samsung.com>
Mon, 3 Mar 2014 02:28:40 +0000 (11:28 +0900)
committerJaegeuk Kim <jaegeuk.kim@samsung.com>
Mon, 3 Mar 2014 02:28:40 +0000 (11:28 +0900)
This patch fixes performance regression of dbench reported by
Alex <hbx7d@yandex.com>.

This issue was revealed by Phoronix tests results:
http://www.phoronix.com/scan.php?page=article&item=linux_314_ssdfs&num=2

It turns out that we need to assign WRITE_SYNC to the node writes, if
fsync is triggered.

The performance numbers are like below, which is measured by Alex.
1. 355MB/s       ext4
2. 225MB/s       f2fs : WRITE for node writes
3. 525MB/s       f2fs : WRITE_SYNC for node writes

Reported-And-Tested-by: Alex <hbx7d@yandex.com>.
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
fs/f2fs/file.c

index 00f937ec4794b289a7f2ccfb9320b8a34a744522..a4cc1d6bdd848f8f402d02286ecbd140855223d2 100644 (file)
@@ -115,7 +115,7 @@ int f2fs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
        int ret = 0;
        bool need_cp = false;
        struct writeback_control wbc = {
-               .sync_mode = WB_SYNC_NONE,
+               .sync_mode = WB_SYNC_ALL,
                .nr_to_write = LONG_MAX,
                .for_reclaim = 0,
        };
This page took 0.028268 seconds and 5 git commands to generate.