|
From: Szakacsits S. <sz...@si...> - 2004-03-21 22:02:26
|
On Sun, 21 Mar 2004, Anton Altaparmakov wrote:
> On Thu, 18 Mar 2004, Szakacsits Szabolcs wrote:
> > I've made some cloning tests with SystemRescueCD 0.2.12_rc1.
> Is this a 2.4 or 2.6 kernel? (Just curious.)
The Gentoo based SystemRescueCD uses 2.4.25, patched with the rewritten
NTFS driver.
> > Source and destination partitions were 10 GB. Source had 5 GB data
> > scattered around the disk. Destination was empty. Both partitions were on
> > the same disk. In case of NTFS, I losetup a big file on a rw mounted NTFS.
> > 'raw' is a plain cloning between two partitions (so no filesystem on the
> > destination partition). Results,
> >
> > ----------------------------------------
> > dest fs user system CPU% real
> > ----------------------------------------
> > raw 1.55 15:02 3 9:04.56
> > ext2 1.75 18.16 3 9:11.25
> > ext3 1.50 21.51 4 9:23.19
> > XFS 1.63 21.19 3 9:31.92
> > NTFS 1.61 49.85 8 9:40.18
> > ReiserFS 1.65 24.83 4 10:10.39
> > JFS 1.82 29.87 4 12:30.33
>
> That is quite intersting!
Yes. For me three very interesting things:
- NTFS being (much) faster than JFS and Reiserfs. But please note,
NTFS used an already preallocated file and the others didn't have
this advantage (but they didn't have the loopback overhead).
- Filesystems are useless from performance point of view :-)
All of them failed to be faster than the raw disk cloning!
Why? I don't know but they should have optimised away the disk
seeks. Instead they introduced such a big overhead that the
cloning operation became slower than just plain device cloning.
- The reason JFS being much slower than others was that it constantly
changed two IO scheduling algoriths. One of them was what others used
(basically read a lot then write a lot) and the other had extremely
bad throughput (read and write at the same time, the seeks killed
the throughput).
The background why I made this quick test:
http://www.sysresccd.org/forums/viewtopic.php?t=242
> I wonder how much better NTFS would do if the other fs were running
> over loopmount, too.
Me too but I don't have the time for it right now. But perhaps I'll make a
script that does these automatically so I could schedule it also to run
when I'm away from computers.
I also wanted to test
- NTFS without losetup, with a specially compiled ntfsclone. I made
the code but forgot to run the test :(
- Run it on XP also. But ntfsclone doesn't work. If somebody knows
how to use it on XP please tell me :)
When I tried to clone to a file ftruncate failed. When I tried to
clone to a device I got permission denied. Also the displayed size
values were wrong, e.g. -12313213124312. It was precompiled 1.9.0
from the official(?) cygwin repository.
Szaka
|
