Directory layout preference improvements from Grigoriy Orlov <gluk@ptci.ru>.

His description of the problem and solution follow. My own tests show
speedups on typical filesystem intensive workloads of 5% to 12% which
is very impressive considering the small amount of code change involved.

------

  One day I noticed that some file operations run much faster on
small file systems then on big ones. I've looked at the ffs
algorithms, thought about them, and redesigned the dirpref algorithm.

  First I want to describe the results of my tests. These results are old
and I have improved the algorithm after these tests were done. Nevertheless
they show how big the perfomance speedup may be. I have done two file/directory
intensive tests on a two OpenBSD systems with old and new dirpref algorithm.
The first test is "tar -xzf ports.tar.gz", the second is "rm -rf ports".
The ports.tar.gz file is the ports collection from the OpenBSD 2.8 release.
It contains 6596 directories and 13868 files. The test systems are:

1. Celeron-450, 128Mb, two IDE drives, the system at wd0, file system for
   test is at wd1. Size of test file system is 8 Gb, number of cg=991,
   size of cg is 8m, block size = 8k, fragment size = 1k OpenBSD-current
   from Dec 2000 with BUFCACHEPERCENT=35

2. PIII-600, 128Mb, two IBM DTLA-307045 IDE drives at i815e, the system
   at wd0, file system for test is at wd1. Size of test file system is 40 Gb,
   number of cg=5324, size of cg is 8m, block size = 8k, fragment size = 1k
   OpenBSD-current from Dec 2000 with BUFCACHEPERCENT=50

You can get more info about the test systems and methods at:
http://www.ptci.ru/gluk/dirpref/old/dirpref.html

                              Test Results

             tar -xzf ports.tar.gz               rm -rf ports
  mode  old dirpref new dirpref speedup old dirprefnew dirpref speedup
                             First system
 normal     667         472      1.41       477        331       1.44
 async      285         144      1.98       130         14       9.29
 sync       768         616      1.25       477        334       1.43
 softdep    413         252      1.64       241         38       6.34
                             Second system
 normal     329         81       4.06       263.5       93.5     2.81
 async      302         25.7    11.75       112          2.26   49.56
 sync       281         57.0     4.93       263         90.5     2.9
 softdep    341         40.6     8.4        284          4.76   59.66

"old dirpref" and "new dirpref" columns give a test time in seconds.
speedup - speed increasement in times, ie. old dirpref / new dirpref.

------

Algorithm description

The old dirpref algorithm is described in comments:

/*
 * Find a cylinder to place a directory.
 *
 * The policy implemented by this algorithm is to select from
 * among those cylinder groups with above the average number of
 * free inodes, the one with the smallest number of directories.
 */

A new directory is allocated in a different cylinder groups than its
parent directory resulting in a directory tree that is spreaded across
all the cylinder groups. This spreading out results in a non-optimal
access to the directories and files. When we have a small filesystem
it is not a problem but when the filesystem is big then perfomance
degradation becomes very apparent.

What I mean by a big file system ?

  1. A big filesystem is a filesystem which occupy 20-30 or more percent
     of total drive space, i.e. first and last cylinder are physically
     located relatively far from each other.
  2. It has a relatively large number of cylinder groups, for example
     more cylinder groups than 50% of the buffers in the buffer cache.

The first results in long access times, while the second results in
many buffers being used by metadata operations. Such operations use
cylinder group blocks and on-disk inode blocks. The cylinder group
block (fs->fs_cblkno) contains struct cg, inode and block bit maps.
It is 2k in size for the default filesystem parameters. If new and
parent directories are located in different cylinder groups then the
system performs more input/output operations and uses more buffers.
On filesystems with many cylinder groups, lots of cache buffers are
used for metadata operations.

My solution for this problem is very simple. I allocate many directories
in one cylinder group. I also do some things, so that the new allocation
method does not cause excessive fragmentation and all directory inodes
will not be located at a location far from its file's inodes and data.
The algorithm is:
/*
 * Find a cylinder group to place a directory.
 *
 * The policy implemented by this algorithm is to allocate a
 * directory inode in the same cylinder group as its parent
 * directory, but also to reserve space for its files inodes
 * and data. Restrict the number of directories which may be
 * allocated one after another in the same cylinder group
 * without intervening allocation of files.
 *
 * If we allocate a first level directory then force allocation
 * in another cylinder group.
 */

  My early versions of dirpref give me a good results for a wide range of
file operations and different filesystem capacities except one case:
those applications that create their entire directory structure first
and only later fill this structure with files.

  My solution for such and similar cases is to limit a number of
directories which may be created one after another in the same cylinder
group without intervening file creations. For this purpose, I allocate
an array of counters at mount time. This array is linked to the superblock
fs->fs_contigdirs[cg]. Each time a directory is created the counter
increases and each time a file is created the counter decreases. A 60Gb
filesystem with 8mb/cg requires 10kb of memory for the counters array.

  The maxcontigdirs is a maximum number of directories which may be created
without an intervening file creation. I found in my tests that the best
performance occurs when I restrict the number of directories in one cylinder
group such that all its files may be located in the same cylinder group.
There may be some deterioration in performance if all the file inodes
are in the same cylinder group as its containing directory, but their
data partially resides in a different cylinder group. The maxcontigdirs
value is calculated to try to prevent this condition. Since there is
no way to know how many files and directories will be allocated later
I added two optimization parameters in superblock/tunefs. They are:

        int32_t  fs_avgfilesize;   /* expected average file size */
        int32_t  fs_avgfpdir;      /* expected # of files per directory */

These parameters have reasonable defaults but may be tweeked for special
uses of a filesystem. They are only necessary in rare cases like better
tuning a filesystem being used to store a squid cache.

I have been using this algorithm for about 3 months. I have done
a lot of testing on filesystems with different capacities, average
filesize, average number of files per directory, and so on. I think
this algorithm has no negative impact on filesystem perfomance. It
works better than the default one in all cases. The new dirpref
will greatly improve untarring/removing/coping of big directories,
decrease load on cvs servers and much more. The new dirpref doesn't
speedup a compilation process, but also doesn't slow it down.

Obtained from:	Grigoriy Orlov <gluk@ptci.ru>

3 files changed, 143 insertions, 21 deletions

diff --git a/sys/ufs/ffs/ffs_alloc.c b/sys/ufs/ffs/ffs_alloc.c
index 9476933..81fb75e 100644
--- a/sys/ufs/ffs/ffs_alloc.c
+++ b/sys/ufs/ffs/ffs_alloc.c
@@ -71,7 +71,7 @@ static void	ffs_clusteracct	__P((struct fs *, struct cg *, ufs_daddr_t,
 				     int));
 static ufs_daddr_t ffs_clusteralloc __P((struct inode *, int, ufs_daddr_t,
 	    int));
-static ino_t	ffs_dirpref __P((struct fs *));
+static ino_t	ffs_dirpref __P((struct inode *));
 static ufs_daddr_t ffs_fragextend __P((struct inode *, int, long, int, int));
 static void	ffs_fserr __P((struct fs *, u_int, char *));
 static u_long	ffs_hashalloc
@@ -593,12 +593,23 @@ ffs_valloc(pvp, mode, cred, vpp)
 		goto noinodes;
 
 	if ((mode & IFMT) == IFDIR)
-		ipref = ffs_dirpref(fs);
+		ipref = ffs_dirpref(pip);
 	else
 		ipref = pip->i_number;
 	if (ipref >= fs->fs_ncg * fs->fs_ipg)
 		ipref = 0;
 	cg = ino_to_cg(fs, ipref);
+	/*
+	 * Track number of dirs created one after another
+	 * in a same cg without intervening by files.
+	 */
+	if ((mode & IFMT) == IFDIR) {
+		if (fs->fs_contigdirs[cg] < 255)
+			fs->fs_contigdirs[cg]++;
+	} else {
+		if (fs->fs_contigdirs[cg] > 0)
+			fs->fs_contigdirs[cg]--;
+	}
 	ino = (ino_t)ffs_hashalloc(pip, cg, (long)ipref, mode,
 					(allocfcn_t *)ffs_nodealloccg);
 	if (ino == 0)
@@ -633,28 +644,112 @@ noinodes:
 }
 
 /*
- * Find a cylinder to place a directory.
+ * Find a cylinder group to place a directory.
+ *
+ * The policy implemented by this algorithm is to allocate a
+ * directory inode in the same cylinder group as its parent
+ * directory, but also to reserve space for its files inodes
+ * and data. Restrict the number of directories which may be
+ * allocated one after another in the same cylinder group
+ * without intervening allocation of files.
  *
- * The policy implemented by this algorithm is to select from
- * among those cylinder groups with above the average number of
- * free inodes, the one with the smallest number of directories.
+ * If we allocate a first level directory then force allocation
+ * in another cylinder group.
  */
 static ino_t
-ffs_dirpref(fs)
-	register struct fs *fs;
+ffs_dirpref(pip)
+	struct inode *pip;
 {
-	int cg, minndir, mincg, avgifree;
+	register struct fs *fs;
+	int cg, prefcg, dirsize, cgsize;
+	int avgifree, avgbfree, avgndir, curdirsize;
+	int minifree, minbfree, maxndir;
+	int mincg, minndir;
+	int maxcontigdirs;
+
+	fs = pip->i_fs;
 
 	avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
-	minndir = fs->fs_ipg;
-	mincg = 0;
-	for (cg = 0; cg < fs->fs_ncg; cg++)
-		if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
-		    fs->fs_cs(fs, cg).cs_nifree >= avgifree) {
-			mincg = cg;
-			minndir = fs->fs_cs(fs, cg).cs_ndir;
+	avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
+	avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg;
+
+	/*
+	 * Force allocation in another cg if creating a first level dir.
+	 */
+	if (ITOV(pip)->v_flag & VROOT) {
+		prefcg = arc4random() % fs->fs_ncg;
+		mincg = prefcg;
+		minndir = fs->fs_ipg;
+		for (cg = prefcg; cg < fs->fs_ncg; cg++)
+			if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
+			    fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
+			    fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
+				mincg = cg;
+				minndir = fs->fs_cs(fs, cg).cs_ndir;
+			}
+		for (cg = 0; cg < prefcg; cg++)
+			if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
+			    fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
+			    fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
+				mincg = cg;
+				minndir = fs->fs_cs(fs, cg).cs_ndir;
+			}
+		return ((ino_t)(fs->fs_ipg * mincg));
+	}
+
+	/*
+	 * Count various limits which used for
+	 * optimal allocation of a directory inode.
+	 */
+	maxndir = min(avgndir + fs->fs_ipg / 16, fs->fs_ipg);
+	minifree = avgifree - fs->fs_ipg / 4;
+	if (minifree < 0)
+		minifree = 0;
+	minbfree = avgbfree - fs->fs_fpg / fs->fs_frag / 4;
+	if (minbfree < 0)
+		minbfree = 0;
+	cgsize = fs->fs_fsize * fs->fs_fpg;
+	dirsize = fs->fs_avgfilesize * fs->fs_avgfpdir;
+	curdirsize = avgndir ? (cgsize - avgbfree * fs->fs_bsize) / avgndir : 0;
+	if (dirsize < curdirsize)
+		dirsize = curdirsize;
+	maxcontigdirs = min(cgsize / dirsize, 255);
+	if (fs->fs_avgfpdir > 0)
+		maxcontigdirs = min(maxcontigdirs,
+				    fs->fs_ipg / fs->fs_avgfpdir);
+	if (maxcontigdirs == 0)
+		maxcontigdirs = 1;
+
+	/*
+	 * Limit number of dirs in one cg and reserve space for 
+	 * regular files, but only if we have no deficit in
+	 * inodes or space.
+	 */
+	prefcg = ino_to_cg(fs, pip->i_number);
+	for (cg = prefcg; cg < fs->fs_ncg; cg++)
+		if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
+		    fs->fs_cs(fs, cg).cs_nifree >= minifree &&
+	    	    fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
+			if (fs->fs_contigdirs[cg] < maxcontigdirs)
+				return ((ino_t)(fs->fs_ipg * cg));
+		}
+	for (cg = 0; cg < prefcg; cg++)
+		if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
+		    fs->fs_cs(fs, cg).cs_nifree >= minifree &&
+	    	    fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
+			if (fs->fs_contigdirs[cg] < maxcontigdirs)
+				return ((ino_t)(fs->fs_ipg * cg));
 		}
-	return ((ino_t)(fs->fs_ipg * mincg));
+	/*
+	 * This is a backstop when we have deficit in space.
+	 */
+	for (cg = prefcg; cg < fs->fs_ncg; cg++)
+		if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
+			return ((ino_t)(fs->fs_ipg * cg));
+	for (cg = 0; cg < prefcg; cg++)
+		if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
+			break;
+	return ((ino_t)(fs->fs_ipg * cg));
 }
 
 /*
diff --git a/sys/ufs/ffs/ffs_vfsops.c b/sys/ufs/ffs/ffs_vfsops.c
index 9803a22..17fa431 100644
--- a/sys/ufs/ffs/ffs_vfsops.c
+++ b/sys/ufs/ffs/ffs_vfsops.c
@@ -624,6 +624,7 @@ ffs_mountfs(devvp, mp, p, malloctype)
 	blks = howmany(size, fs->fs_fsize);
 	if (fs->fs_contigsumsize > 0)
 		size += fs->fs_ncg * sizeof(int32_t);
+	size += fs->fs_ncg * sizeof(u_int8_t);
 	space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
 	fs->fs_csp = space;
 	for (i = 0; i < blks; i += fs->fs_frag) {
@@ -645,6 +646,15 @@ ffs_mountfs(devvp, mp, p, malloctype)
 		for (i = 0; i < fs->fs_ncg; i++)
 			*lp++ = fs->fs_contigsumsize;
 	}
+	size = fs->fs_ncg * sizeof(u_int8_t);
+	fs->fs_contigdirs = (u_int8_t *)space;
+	space = (u_int8_t *)space + size;
+	bzero(fs->fs_contigdirs, size);
+	/* Compatibility for old filesystems 	   XXX */
+	if (fs->fs_avgfilesize <= 0)		/* XXX */
+		fs->fs_avgfilesize = AVFILESIZ;	/* XXX */
+	if (fs->fs_avgfpdir <= 0)		/* XXX */
+		fs->fs_avgfpdir = AFPDIR;	/* XXX */
 	mp->mnt_data = (qaddr_t)ump;
 	mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
 	mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
diff --git a/sys/ufs/ffs/fs.h b/sys/ufs/ffs/fs.h
index 4083c14..c11a9f8 100644
--- a/sys/ufs/ffs/fs.h
+++ b/sys/ufs/ffs/fs.h
@@ -108,15 +108,17 @@
 /*
  * There is a 128-byte region in the superblock reserved for in-core
  * pointers to summary information. Originally this included an array
- * of pointers to blocks of struct csum; now there are just two
+ * of pointers to blocks of struct csum; now there are just three
  * pointers and the remaining space is padded with fs_ocsp[].
  *
  * NOCSPTRS determines the size of this padding. One pointer (fs_csp)
  * is taken away to point to a contiguous array of struct csum for
  * all cylinder groups; a second (fs_maxcluster) points to an array
- * of cluster sizes that is computed as cylinder groups are inspected.
+ * of cluster sizes that is computed as cylinder groups are inspected,
+ * and the third points to an array that tracks the creation of new
+ * directories.
  */
-#define	NOCSPTRS	((128 / sizeof(void *)) - 2)
+#define	NOCSPTRS	((128 / sizeof(void *)) - 3)
 
 /*
  * A summary of contiguous blocks of various sizes is maintained
@@ -142,6 +144,18 @@
 #define DEFAULTOPT	FS_OPTTIME
 
 /*
+ * Grigoriy Orlov <gluk@ptci.ru> has done some extensive work to fine
+ * tune the layout preferences for directories within a filesystem.
+ * His algorithm can be tuned by adjusting the following parameters
+ * which tell the system the average file size and the average number
+ * of files per directory. These defaults are well selected for typical
+ * filesystems, but may need to be tuned for odd cases like filesystems
+ * being used for sqiud caches or news spools.
+ */
+#define AVFILESIZ	16384	/* expected average file size */
+#define AFPDIR		64	/* expected number of files per directory */
+
+/*
  * The maximum number of snapshot nodes that can be associated
  * with each filesystem. This limit affects only the number of
  * snapshot files that can be recorded within the superblock so
@@ -273,12 +287,15 @@ struct fs {
 /* these fields retain the current block allocation info */
 	int32_t	 fs_cgrotor;		/* last cg searched */
 	void 	*fs_ocsp[NOCSPTRS];	/* padding; was list of fs_cs buffers */
+	u_int8_t *fs_contigdirs;	/* # of contiguously allocated dirs */
 	struct csum *fs_csp;		/* cg summary info buffer for fs_cs */
 	int32_t	*fs_maxcluster;		/* max cluster in each cyl group */
 	int32_t	 fs_cpc;		/* cyl per cycle in postbl */
 	int16_t	 fs_opostbl[16][8];	/* old rotation block list head */
 	int32_t	 fs_snapinum[FSMAXSNAP];/* list of snapshot inode numbers */
-	int32_t	 fs_sparecon[30];	/* reserved for future constants */
+	int32_t	 fs_avgfilesize;	/* expected average file size */
+	int32_t	 fs_avgfpdir;		/* expected # of files per directory */
+	int32_t	 fs_sparecon[28];	/* reserved for future constants */
 	int32_t	 fs_contigsumsize;	/* size of cluster summary array */ 
 	int32_t	 fs_maxsymlinklen;	/* max length of an internal symlink */
 	int32_t	 fs_inodefmt;		/* format of on-disk inodes */