[Ocfs2-tools-devel] [PATCH 06/11] libocfs2: Increase the buffer size in ocfs2_open_inode_scan()

[Sunil Mushran]

ocfs2_open_inode_scan() is used by utilities that are looking to read all
the inodes. This patch increases the buffer from one cluster to 4MB to improve
the performance of pass 1 in fsck.

The test was run on a 2TB volume having 15M files. The cache size was 820MB.

Pass 1: Checking inodes and blocks

Before:
  I/O read disk/cache: 64698MB / 66MB, write: 0MB, rate: 17.15MB/s
  Times real: 3775.640s, user: 218.995s, sys: 55.703s

After:
  I/O read disk/cache: 64698MB / 66MB, write: 0MB, rate: 37.28MB/s
  Times real: 1737.086s, user: 193.295s, sys: 25.405s

Signed-off-by: Sunil Mushran <sunil.mushran at oracle.com>
---
 libocfs2/inode_scan.c |   18 +++++++-----------
 1 files changed, 7 insertions(+), 11 deletions(-)

diff --git a/libocfs2/inode_scan.c b/libocfs2/inode_scan.c
index d215f56..317fc64 100644
--- a/libocfs2/inode_scan.c
+++ b/libocfs2/inode_scan.c
@@ -331,17 +331,13 @@ errcode_t ocfs2_open_inode_scan(ocfs2_filesys *fs,
 	if (ret)
 		goto out_scan;
 
-	/* Minimum 8 inodes in the buffer */
-	scan->buffer_blocks = fs->fs_clustersize / fs->fs_blocksize;
-	if (scan->buffer_blocks < 8) {
-		scan->buffer_blocks =
-			((8 * fs->fs_blocksize) +
-			 (fs->fs_clustersize - 1)) /
-			fs->fs_clustersize;
-		scan->buffer_blocks =
-			ocfs2_clusters_to_blocks(fs,
-						 scan->buffer_blocks);
-	}
+	/*
+	 * Ideally the buffer size should be one cpg. But finding that value
+	 * is not worth the effort. Instead we default to 4MB, which is a
+	 * typical value in most ocfs2 file systems.
+	 */
+#define OPEN_SCAN_BUFFER_SIZE	(4 * 1024 * 1024)
+	scan->buffer_blocks = OPEN_SCAN_BUFFER_SIZE / fs->fs_blocksize;
 
 	ret = ocfs2_malloc_blocks(fs->fs_io, scan->buffer_blocks,
 				  &scan->group_buffer);
-- 
1.7.4.1