0. Benchmark system:
Host: IBM Thinkpad T40, 1.5GHz Pentium M, 2GiB RAM
Guest: QEMU, 384MiB RAM
	128MiB /dev/hda ext3 rootfs
	512MiB /dev/hdb for LogFS benchmark
	512MiB /dev/hdc for JFFS2 benchmark

Both hdb and hdc contain all 0xff before each benchmark run.  QEMU is
rebootet before each run.  All 1.5GiB for test system are cached in
host memory.

Both LogFS and JFFS2 use an erasesize of 128KiB.  JFFS2 code is stock
2.6.20 without summary, LogFS is current development code.  Block2mtd
had three fixes applied, see
http://git.kernel.org/?p=linux/kernel/git/joern/misc.git;a=log


1. mkfs benchmark

Both LogFS and JFFS2 will create a new filesystem when mounting an
erased mtd partition.  JFFS2 background thread will erase all blocks
and write cleanmarkers to them.  Umount must wait for this.  Times are
wallclock times; highest, median and lowest of three runs.

Mount LogFS
real	0m0.179s
real	0m0.171s
real	0m0.169s
Mount JFFS2
real	0m4.841s
real	0m4.715s
real	0m4.567s
Umount LogFS
real	0m0.100s
real	0m0.081s
real	0m0.079s
Umount JFFS2
real	2m31.043s
real	2m28.684s
real	2m27.144s


2. Mount/umount benchmark

Mount and umount filesystem.  Times are combined wallclock times for
mount and umount; highest, median and lowest of 12 runs.

Mount/umount LogFS
real	0m0.308s
real	0m0.203s
real	0m0.184s

Mount/umount JFFS2
real	0m2.312s
real	0m2.173s
real	0m2.095s


3. Write benchmark

Mount a filesystem, copy a 16MiB random-data file to it, umount the
filesystem again.  Times are combined wallclock times for mount, copy
and umount; highest, median and lowest of 12 runs.

Copy 16MiB file to LogFS
real	0m35.871s
real	0m25.046s
real	0m21.614s

Copy 16MiB file to JFFS2
real	0m36.931s
real	0m36.274s
real	0m33.525s


4. Read benchmark

Mount a filesystem, read a 16MiB random-data file from it, umount the
filesystem again.  Times are combined wallclock times for mount, read
and umount; highest, median and lowest of 12 runs.

Read 16MiB from LogFS
real	0m2.290s
real	0m1.771s
real	0m1.625s

Read 16MiB from JFFS2
real	0m7.114s
real	0m6.701s
real	0m6.227s


5. Comments

There are lies, damn lies and benchmarks.  In particular, this
benchmark is comparing apples to oranges.  JFFS2 has compression
support and will attempt to compress data on writes.  Benchmark data
is random, so compression never works and uncompressed data has to be
written instead.  LogFS does not have compression support yet, so it
will write uncompressed data without wasting any time.

The test device is effectively RAM of the host system.  LogFS
currently has to write more to the device than JFFS2, but this
disadvantage is hidden by the fast test device.

JFFS2 mounts are slow, on average 2s slower for an empty device.
Substracting mount times from read and write times improves the JFFS2
numbers a little.

Overall the write times give rough estimates at best due to above
reasons.  Read times are more meaningful and clearly favor LogFS.