On Tue, 25 Jan 2005, Michael M�ller wrote:

> Hi Michael!

> If I understand right what you are looking for you could give
> Reiser-FS a try. I have one partition with Reiser. It stores my copy
> of the freecddb files: many many small text files.

I think he's looking for "dense" file support, which is like an mount 
option "nosparse" that doesn't exist. I would guess that the major 
selling point of "dense" file support is that once a file has been 
allocated, you can be guaranteed that all of the space necessary to 
store the file exists on disk.

According to my tests, reiser actually does the worst of every 
filesystem I tested (with respect to overhead).  reiserv4 does the best, 
with ext2/3 coming in a close second.  I'm sure my test is flawed with 
only a 100MB "partition" to test with, as the default blocksizes, etc... 
are likely to be different on larger filesystems.

This is what I did. I have 410 very small files (average size 
close to 1KB, most are less than that, with a very small number 
larger than 3K.)

I made a new logical volume of 100MB, and formatted it with JFS, copied 
over the files, and ran:

report disk usage in BLOCKS:
  du -s:  1684    
report *actual* disk usage in BLOCKS
  du -s --apparent-size: 293
report disk usage in bytes
  du -s --block-size=1: 1724416
report *actual* disk usage in bytesk
  du -s --block-size=1 --apparent-size: 299600

1724416/299600 = 575% overhead.

I repeated with other filesysems.  In a table:

  fs               -s  -s+as     -s +bs   -s+bs+as   overhead
  -----------------------------------------------------------
  jfs:           1684    293    1724416     299600   575%
  reiser (v3):   1536    243    1572864     247888   634%
  ext3:           478    260     489472     265968    84% 
  xfs:           1588    289    1626112     295916   549%
  reiser4:        346    229     353792     233671    51%

As you can see, reiser4 had the least overhead, followed by ext3, jfs,
xfs, then reiserv3. 

What is the cause of the overhead, or am I wrong entirely about it even
*being* overhead?

--
Jon Nelson <[EMAIL PROTECTED]>

Reply via email to