freebsd-skq/sys/ufs/ffs
Matthew Dillon 1b7e3dafdf Fix a file-rewrite performance case for UFS[2]. When rewriting portions
of a file in chunks that are less then the filesystem block size, if the
data is not already cached the system will perform a read-before-write.
The problem is that it does this on a block-by-block basis, breaking up the
I/Os and making clustering impossible for the writes.  Programs such
as INN using cyclic file buffers suffer greatly.  This problem is only going
to get worse as we use larger and larger filesystem block sizes.

The solution is to extend the sequential heuristic so UFS[2] can perform
a far larger read and readahead when dealing with this case.

(note: maximum disk write bandwidth is 27MB/sec thru filesystem)
(note: filesystem blocksize in test is 8K (1K frag))
dd if=/dev/zero of=test.dat bs=1k count=2m conv=notrunc

Before:  (note half of these are reads)
      tty             da0              da1             acd0             cpu
 tin tout  KB/t tps  MB/s   KB/t tps  MB/s   KB/t tps  MB/s  us ni sy in id
   0   76 14.21 598  8.30   0.00   0  0.00   0.00   0  0.00   0  0  7  1 92
   0   76 14.09 813 11.19   0.00   0  0.00   0.00   0  0.00   0  0  9  5 86
   0   76 14.28 821 11.45   0.00   0  0.00   0.00   0  0.00   0  0  8  1 91

After:	(note half of these are reads)
      tty             da0              da1             acd0             cpu
 tin tout  KB/t tps  MB/s   KB/t tps  MB/s   KB/t tps  MB/s  us ni sy in id
   0   76 63.62 434 26.99   0.00   0  0.00   0.00   0  0.00   0  0 18  1 80
   0   76 63.58 424 26.30   0.00   0  0.00   0.00   0  0.00   0  0 17  2 82
   0   76 63.82 438 27.32   0.00   0  0.00   0.00   0  0.00   1  0 19  2 79

Reviewed by:	mckusick
Approved by:	re
X-MFC after:	immediately (was heavily tested in -stable for 4 months)
2002-10-18 22:52:41 +00:00
..
ffs_alloc.c intmax_t is printed with %jd, not %lld. 2002-09-19 03:55:30 +00:00
ffs_balloc.c Fix a file-rewrite performance case for UFS[2]. When rewriting portions 2002-10-18 22:52:41 +00:00
ffs_extern.h Introduce typedefs for the member functions of struct vfsops and employ 2002-08-13 10:05:50 +00:00
ffs_inode.c - Convert locks to use standard macros. 2002-09-25 02:49:48 +00:00
ffs_snapshot.c Change locking so that all snapshots on a particular filesystem share 2002-10-16 00:19:23 +00:00
ffs_softdep_stub.c Add a missing argument to the stub for softdep_setup_freeblocks. 2002-07-20 04:07:15 +00:00
ffs_softdep.c When spamming me with a printf(9), under DIAGNOSTIC, at least be nice enough 2002-09-28 19:04:49 +00:00
ffs_subr.c I forgot this bit of uglyness in the fsck_ffs cleanup. 2002-07-31 07:01:18 +00:00
ffs_tables.c This commit adds basic support for the UFS2 filesystem. The UFS2 2002-06-21 06:18:05 +00:00
ffs_vfsops.c If the FS_MULTILABEL flag is set in a UFS or UFS2 superblock, 2002-10-15 20:00:06 +00:00
ffs_vnops.c Fix a file-rewrite performance case for UFS[2]. When rewriting portions 2002-10-18 22:52:41 +00:00
fs.h Define two new superblock file system flags: 2002-10-14 17:07:11 +00:00
README.snapshot Fix a type: s/your are/you are/ 2002-07-12 19:56:31 +00:00
README.softupdates Update to reflect current status. 2000-07-08 02:31:21 +00:00
softdep.h Add support to UFS2 to provide storage for extended attributes. 2002-07-19 07:29:39 +00:00

$FreeBSD$

Using Soft Updates

To enable the soft updates feature in your kernel, add option
SOFTUPDATES to your kernel configuration.

Once you are running a kernel with soft update support, you need to enable
it for whichever filesystems you wish to run with the soft update policy.
This is done with the -n option to tunefs(8) on the UNMOUNTED filesystems,
e.g. from single-user mode you'd do something like:

	tunefs -n enable /usr

To permanently enable soft updates on the /usr filesystem (or at least
until a corresponding ``tunefs -n disable'' is done).


Soft Updates Copyright Restrictions

As of June 2000 the restrictive copyright has been removed and 
replaced with a `Berkeley-style' copyright. The files implementing
soft updates now reside in the sys/ufs/ffs directory and are
compiled into the generic kernel by default.


Soft Updates Status

The soft updates code has been running in production on many
systems for the past two years generally quite successfully.
The two current sets of shortcomings are:

1) On filesystems that are chronically full, the two minute lag
   from the time a file is deleted until its free space shows up
   will result in premature filesystem full failures. This
   failure mode is most evident in small filesystems such as
   the root. For this reason, use of soft updates is not
   recommended on the root filesystem.

2) If your system routines runs parallel processes each of which
   remove many files, the kernel memory rate limiting code may
   not be able to slow removal operations to a level sustainable
   by the disk subsystem. The result is that the kernel runs out
   of memory and hangs.

Both of these problems are being addressed, but have not yet
been resolved. There are no other known problems at this time.


How Soft Updates Work

For more general information on soft updates, please see:
	http://www.mckusick.com/softdep/
	http://www.ece.cmu.edu/~ganger/papers/CSE-TR-254-95/

--
Marshall Kirk McKusick <mckusick@mckusick.com>
July 2000