wow! Maybe we've got something to brag about in GNU/Linux
 

Hello.

I was bored today and decided to conduct a few benchmarks on my Lenovo 3000 g530 laptop with Linux/Vista. I wanted to compare their copy speeds from a couple different filesystems. I discovered that ntfs writing is even FASTER under Gnu/Linux, and even faster on ext2 filesystem vs. ntfs filesystem. What I did was copy the linux kernel source directory (roughly 478mb and 33000plus files(object files, etc.)) in each OS and from ntfs->ntfs on Vista/ then on Gnu/linux from ntfs->ntfs, ntfs->ext2, and ext2->ext2. I found:

from ntfs->ntfs on Vista took about 7.5 minutes.

from ntfs->ntfs on Gnu/Linux took about 5minutes 45 seconds.

from ntfs->ext2 on Gnu/Linux took about 3minutes 32 seconds.

from ext2->ext2 on Gnu/Linux took 20 seconds to copy 33000 files, 478mb!! Wow!

So, I'd say that ext2 is WAY faster than ntfs and that ntfs writes are faster under linux than windows. I did notice that the size of the directory created on the ext2 filesystem is slightly larger (510Mb) though.

Linux wins on copy speeds!! :)

The last test may not be valid. If you copied the files from one directory to another on the same filesystem, then only the directory entries were changed but the inodes and files where not moved.

ah, that would explain why it was so fast. But Linux still seems faster writing to ntfs than windows. Now if only I could get usb 2.0 to work as fast in Linux as it does under windows, than we'd have a sure winner. :)

I also think that you cannot compare a journaled filesystem (NTFS) to a non-journaled filesystem (ext2).

Still, you should know that NTFS is not a performance filesystem. But then is anything related to M$ about performance, they don't even belong in the same sentence.

Ext3 and Reiserfs were designed with compatibility and durability in mind rather than performance. It might be used on a 386 computer and will be used on workstations where a user is more apt to shut down the computer improperly. Some Linux filesystems are better at serving very large files, in a video server for example, or for large databases. They may not be able to recover as easily. It is expected that they run on servers with power backup.

If this whole test is done on one drive the test is invalid. Speed can also be determined on the amount of distance the head needs to move, how much the head needs to move (is the file system fragmented?), and how much caching is available to move larger chunks at once.

Your fastest speeds will take place on the outer edges of the platter because the surface of the platter is moving faster with respect to the head than if the head were at the center.

You now must do this test all over again. Go out and buy two identical hard drives connected to sata with linux and windows installed onto the third. Format both the drives in NTFS put some test data on one and copy it. Then do it again in Linux. This must be done at least 5 times with different test data sizes and number of files. (for greater resolution you should also test with different block sizes as well.

OK now format one with ext3 since any sane person uses a journaled file system. and repeat the test again in both. Windows will probably win on this one depending on the ext3 driver you have. The only one I've seen is for ext2 which would negate the journal as well as file permissions.

Now format both in ext3 and again.

your most interesting results will be windows ntfs/ntfs against linux ext3/ext3.

your worst results will be windows ext3/ext3 and linux ntfs/ntfs because both operating systems don't properly support the other's filesystems and all test results would be invalid anyway.


Casual speed checks are all fun and dandy but in the end don't mean diddly squat. Unless you know exactly what you are testing, your results will be flawed. The most common mistake that's made is not factoring in variable x, not understanding variable y, and only mentioning that variable z was accounted for but in reality doesn't mean anything because variable y wasn't tested correctly and variable x was completely missing!

OK Now that was a boredom rant. Sorry.

Ok. Here's what I was doing regarding copying to the NTFS partition. I have a shared NTFS partition on my laptop (Vista/Linux). I copied the linux source directory (478 or so Mb) to this partition from windows C drive to this shared partition and recorded the time it took (If I remember it was roughly 7.5 minutes) and then I rebooted into Linux and did the same thing and Linux copied the same directory to that shared partition in a little over 5 minutes. So, all I was saying that in seems that Linux copies to NTFS faster than windows.