Hi, can you make two bugreports with "enhancement" status out of this mail please ? bye adrian On Friday 11 November 2005 11:13, jsiebert@arcor.de wrote:
Hi there,
i've been using ext3 as my favorite filesystem for all of my systems for some time now because of stability reasons (had some problems with xfs, reiserfs burns to many cpu cycles). Performance of ext3 in it's standard configuration on openSUSE/SUSE Linux 10 is not that good compared to reiserfs or xfs especially when it comes to write operations. So I've investigated the issue myself and found some things that could be improved in the future.
1. data=ordered vs. data=writeback on / filesystems.
In ordered jounalling mode of ext3 data is always written to disk prior to the metadata (journal data). This ensures data consistency in most cases. However the performance of this option is not that well. In writeback journalling mode the metadata might get comitted to the journal before the data is actually written to the disk. This might lead to data corruption in some cases (e.g. power-off before the data could be written to disk but the journal data already has been commited). But on the other hand performance of this option is considerably better. The YAST2 installer allows to set the writeback journalling mode but not in the / filesystem. I one specifies the data=writeback option in /etc/fstab the / filesystem is mounted in ordered journalling mode although. To get around this one can use the tune2fs tools with the option '-o journal_data_writeback'. This sets the default mount option for the filesystem with which it gets mounted in writeback journalling mode. The same applies to the 'journal' journalling mode.
So my first suggestion for the YAST2 installer is to call' tune2fs -o journal_data_writeback' for the respective device if a user selects the writeback journalling mode on ext3 for the / filesystem. I've tested this on three different machines without any problems.
2. Enable dir_index feature
ext3 has a feature called 'dir_index' which enables faster lookups in large filesystems by using a technique called hashed b-trees. Currently a user which wants to use this feature has to manually convert the filesystem by using 'tune2fs -O dir_index <device>' and 'fsck.ext3 -fD <device>' which takes a long time to complete on large filesystems. This feature can also be enabled by using 'mkfs.ext3 -O dir_index <device>' at filesystem creation time. The YAST2 installer doesn't offer an option for this.
So my second suggestion is to add an option to the YAST2 installer to enable the dir_index feature at filesystem creation time. I've also tested this option on three different system and noticed some speedups especially on slower systems (my notebook).
That's all for today ;-) Maybe these suggestions could be considered for future openSUSE/SUSE Linux releases.
Greetings,
Machen Sie aus 14 Cent spielend bis zu 100 Euro! Die neue Gaming-Area von Arcor - über 50 Onlinespiele im Angebot. http://www.arcor.de/rd/emf-gaming-1
--------------------------------------------------------------------- To unsubscribe, e-mail: opensuse-factory-unsubscribe@opensuse.org For additional commands, e-mail: opensuse-factory-help@opensuse.org
-- Adrian Schroeter SUSE Linux Products GmbH, Maxfeldstr. 5, 90409 Nuernberg, Germany email: adrian@suse.de