On 2009-12-08, at 09:43, Iavor Stoev wrote:
We use this setup for our backup servers.
We use rsync via SSH using hard links as backup technology; the
backup server is pulling the data from several servers.
The setup is 12x1TB disks in RAID6 128k stripe, using ext3 4k block
+ lvm2 with the journal on Gigabyte I-RAM drive 1GB DDR400.
The server has 8GB RAM.
The journal mode is data.
The journal size is 400MB.
When we moved the journal on the external device we have gained like
20%
performance improvement with our backup.
I'm converting several servers to ext4 to see what will be the
performance improvement for our workload.
Do you have any suggestions regarding the journal size and the
overall file system setup?
You should definitely make sure you create enough inodes (use -i or -
N), and use the flex_bg option (enabled by default for ext4) to
improve metadata performance.
Andreas Dilger wrote:
On 2009-12-07, at 14:46, Iavor Stoev wrote:
I wonder if the Ext3's MAX journal size of 102,400 file system
blocks
has been increased in Ext4.
I'm using 10TB 4k block Ext3 file system with external journal on
Gigabyte I-Ram drive and I'm planning a migration to Ext4 system.
And I wonder if I can increase the journal size over 400MB.
Well, even with ext3 the maximum journal size was only for internal
journals. It was always possible to have larger external journal
devices.
With ext4, the maximum journal size WAS increased, though this is
in fact a mke2fs/tune2fs limit so it is also increased for new ext3
filesystems.
Note that with large journals you are also consuming an equal
amount of RAM as the size of the journal, so don't make it crazy
big. Having a journal on SSD is only really noticable for sync-
happy workloads. It isn't noticably better than using a regular
disk for the external journal if you aren't doing a lot of syncs
(e.g. NFS or email).
I've thought in the past that it might be an interesting hack to
use a huge journal device (say 32GB) with data journaling, and then
have the JBD layer get the data blocks from the journal for
checkpointing to the filesystem instead of keeping the buffers
pinned in RAM. That would would allow blazing metadata workloads,
zero seeking, and then checkpointing in bulk to the
filesystem. ... but unfortunately not something I have time to
test out.
Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.
--
To unsubscribe from this list: send the line "unsubscribe linux-
ext4" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
