One of the most important thing why I started research on this topic is:

1, The file 1.avi(~700MB) is downloaded by aMule, the file-system is Reiser3(usage 70%)
#hdparm -t /dev/hda
/dev/hda:
Timing buffered disk reads: 90 MB in 3.05 seconds = 29.48 MB/sec

2, Before:
#filefrag 1.avi
1.avi: 48044 extents found
#time cat 1.avi>/dev/null
real 3m1.478s
user 0m0.020s
sys 0m2.086s

3, After:
#cp 1.avi 2.avi
#filefrag 2.avi
2.avi: 128 extents found
#time cat 2.avi>/dev/null
real 0m25.329s
user 0m0.012s
sys 0m1.329s

4, Conclusion:
After re-allocation by the file-system itself, file fragments decreased to 1/375, file read performance improved 7 times, which means much less disk-seek during movie playing.

5, Future Works:
I'm going to post some tests so all of you could take it on your own "real-world" file-system.

Your tests will never be worth anything, no matter how dire or repetitive. You cannot be your own reference. It simply doesn't work that way.

So now you're claiming to be Keith A. Smith at Harvard? hahahahahahahahahahaha

I'm trying ignore any future subjective/flame-war posts.
While any constructive/objective posts warmly welcomed.

tmcco

Have you contacted any of your references for peer review? Assuming Keith A. Smith "broke open" this field, what does he have to say?

1, No, I have not, but you have a good idea. I'm also considering paper-lize it.

2, I asked Jens Axboe(a guru wrote Linux scheduler codes) about rkelsen's post #50, he said the link rkelsen provided is inaccurate.

Did you ask him for some accurate information then?

What does Jens think of your 'defrag' tool? Does he have an opinion on how effective it is?

Did you ask him about how file fragmentation affects the efficiency of his scheduler? Or whether he bothers defragmenting his drives?

1, The "Theory" page is updated, making it seems less advocacy/propaganda.

2, I'm keeping contact with Jens, hope his reply will come soon.

After reading those pages:

http://defragfs.sourceforge.net/theory.html
http://defragfs.sourceforge.net/theory1.html
http://defragfs.sourceforge.net/theory2.html
http://defragfs.sourceforge.net/defrag.html

Theodore Ts'o(A guru of Linux file-system and other aspect) has some other more constructive suggestions:

"I see you've done a lot of very careful measurements here. One thing
that is worth noting is that your tests can also be viewed as
modelling what happens within a single project directory in a much
larger filesystem, since most filesystems do use some kind of block
allocation group to try to keep related files (usually defined as "in a
single directory") in the same block allocation group.

So where I'm going with this is that while it is a good idea to think
about defragmenting filesystems, and it while I agree with your
premise that filesystems *do* fragment over time, and that it is an
overstatement to say that all Linux filesystems are fragmentation-free
--- it is better to have the filesystem to be as
fragementation-resistent as possible, rather than try to defragment
the system after the fact, as this is slow and subject to all sorts of
complicated race conditions (i.e., what if the file is in use when you
are trying to defragment it).

There is definitely more work that can be done to make ext3/ext4 more
fragmentation resistant, and in the end my personal belief is that
this is at least as important as defragmenters. That's not to say
that folks shouldn't work on defragmentation technology, but it would
be good if we could come up with benchmarks that would in fact
demonstrate that various defragmentation hueristics that depend on
filenames, or other characteristics that show up in real-life
scenarios, would actually make a difference."

So, basically, Mr. Ts'o has told you the same thing as many of the people who posted in this thread, albeit in a slightly different manner...

1, Definitely Ted reads those pages much more carefully than most of you, and much more experienced than you all. So he would not say something like "the test is sequential! // the test is disk full! // I do not feel performance degradation! // you do not need worry about fragmentation! // fragmentation is FAT-only! // your theory is fake! // your theory is ridiculous! // your theory is ever-worthless! // you claimed yourself to be xxx? // ........"

2, Ted agrees with me that "filesystems *do* fragment over time", while most of you don't.

3, Ted agrees with me that "it is an overstatement to say that all Linux filesystems are fragmentation-free", while most of you don't.

4, Ted agrees with me that "it is a good idea to think about defragmenting filesystems", while most of you don't.

5, Ted agrees with me that "There is definitely more work that can be done to make ext3/ext4 more fragmentation resistant", while most of you don't.

6, Ted just point out technological things & constructive suggestions, while most of you just subjective judgments.

7, Ted does not like some of you(which I do not want point out) posting rude/flame words.

8, I definitely noticed Ted's advice: make the process more "real-world"(file names, directories...). As I said in #44: The test scripts are artificial and should reflect the worst performance degradation(even random-read performance dropped to 50%). For a quick view of your own file-system "fragmentation / performance" degradation, just dump your file-system to another partition and try some random-read scenario against the original and the new.

tmcco

Finding a mentor like Mr. Ts'o was very good. Referring to him by his first name in a public forum does not show the greatest respect (which I believe he has earned). One has to be careful about making blind statements like: and much more experienced than you all. You do not know who any of these posters are or what experience they have. Linus T has been known to visit a forum or two. How do you KNOW that he is not one of the previous posters(I do not think he was)? One also has to be extremely careful interpreting other peoples meaning. It would have been better for you to allow Mr. Ts'o to speak for himself.

This sort of statement is usually an attempt to assume the position of authority of another. It's one thing to do it in private, but quite another to do it in a public forum where you can be ridiculed and generally "laughed off the stage".

It would be acceptable for you to say "I agree with Mr Ts'o when he says xxx". It would also be acceptable for you to quote Mr. Ts'o's published works, as long as your quotations are in context. But, to quote a personal email and then to baldly imply that the authority assumes a position of inferiority to you is really beyond the pale.

As we continue to tell you, do the hard work and the research yourself. If you will stop dropping names as if these people are your peers or students, then you may be taken more seriously. You've dug quite a hole for yourself. I would suggest that a bit of humility and hard work will pave the way out of it.

I did not say that Linux filesystems do not fragment. They do but again not as much as Windows filesystems. I did not say your theory is fake. I said your program is fake. In order to do a good defrag, the addresses that points to data have to re-organized or sorted. Copying files from directory and then to another directory. Finally back to the present directory is wrong way to do defrag.

All filesystems fragment over time some are worst than others.

Linux filesystems have some resistence from fragmentation.

Yes, defragging is ok, but can hurt some operating systems.

I did not say that.

I said your program is fake because it goes in the wrong direction for defragging.

There is always hate in this world.

Dumping is not real defragging. Defragging is organizing scattered data to bring back performance. Your program is fake because it does not get the nitty-gritty of the data to sort through all the addresses that points to the data.

DOS/Windows defrag utility organizes the data based on addresses instead of the fake way moving data from directory and then move it back to the present directory.

I recommend asking several hard drive and utility manufactures about the methods and the proper way of defragging. A said "Filesystem Guru" does not give enough proof.

1, "Ted" is well known as Mr Ts'o's nickname and it is widely used. So do not be serious too much.

2, I did not mean "everybody in the world", but "everybody who post in this thread"(if anybody of you thinking yourself a guru, please point out)