| Author |
Message |
|
|
Post subject: Your ref to configure fstab and grub being stupid.
 Posted: Aug 05, 2008 - 12:20 PM
|
|
Joined: Apr 15, 2008
Posts: 129
Location: Australia
Status: Offline
|
|
damentz
I have no problem with a post explaining how to change the journal tweaks about ext3.
What did you find stupid about fstab and grub tweaks you found online?
2) Leaping ahead are you referring to these online tweaks?
fstab can be tweaked to use noatime or nodiratime and have no fsck by making fstab entry end in 0 0
grub can be tweaked by:
changing timeout to 0 or a few seconds
commenting out gfxmenu
using kernel paramenters AKA cheatcodes to disable certain modules as you do not have the hardware etc
feel free to be as blunt as you like. But I do take exception to your use of the word stupid in dismissing grub so readily.
cheerio |
|
|
| |
|
|
|
 |
|
|
|
Post subject:
Posted: Aug 05, 2008 - 04:14 PM
|
|
Joined: Jan 12, 2008
Posts: 289
Location: Ohio/USA
Status: Offline
|
|
Using an ext3 filesystem on a SSD, such as the one on the Asus Eee PC, is generally not recommended due to the limited duty cycle of such flash memory devices, and the fact that Linux does frequent writing of journal logs and other system logs, thus accelerating SSD wear.
However, you can change the default journal flush "commit" time of every 5 seconds and make it something longer (at the risk of losing more metadata, of course, in a power outage or other crash). Suppose you can stand the risk of a 90-second commit cycle, in conjunction with the "noatime" and "data=writeback" options, to keep a modicum of data security without incurring excessive wear on the SSD, you could do the following (well, maybe Damentz would use his tune2fs command ... ):
1. In /etc/fstab:
Code:
UUID=85a0aeb3-f469-4283-9645-9f10616db234 / ext3 defaults,noatime,errors=remount-ro,data=writeback,commit=90 0 1
2. In /etc/sysctl.conf:
Code:
vm.dirty_writeback_centisecs = 9000
3. In /boot/grub/menu.lst:
Code:
# defoptions=quiet splash rootflags=data=writeback,commit=90 remount=/dev/sdb1 resume2=swap=UUID=e4f864b0-344e-4838-af08-3230293c97a0
Then if you make your /var/log a tmpfs so it doesn't write to the SSD, you can have an ext3 filesystem with far less than the default "wear" of the SSD. |
_________________
Intel X6800 on Intel D975XBX2, Nvidia 9600GT
sidux 2008-02 "Έρεβος" 64-bit on JFS filesystem
Asus Eee PC 4G / 701, sidux 2008-03 "Ουρέα"
|
| |
|
|
|
|
|
|
|
Post subject: Re: ext3 tweaks
Posted: Aug 22, 2008 - 01:09 PM
|
|
Joined: May 28, 2007
Posts: 545
Status: Offline
|
|
|
damentz wrote:
...
Just remember not to have two journal options as default.
...
What do you mean by this? |
|
|
| |
|
|
|
|
|
|
|
Post subject: writeback journal only with telling the kernel to do so
Posted: Nov 01, 2008 - 09:18 PM
|
|
Joined: Apr 29, 2008
Posts: 9
|
|
First of all thanks for the good tips. I found a problem concerning the writeback journal mode of ext3, though.
the tune2fs command is not enough. if you want to enable writeback mode on your root file system, you also have to tell the kernel to do so.
Additionally to tune2fs, I added a line in menu.lst and updated grub.
You can find the information here:
http://ubuntuforums.org/showthread.php?t=107856
and in the man pages of "mount". there it says for a root filesystem in writeback mode you have to tell the kernel. |
|
|
| |
|
|
|
|
|
|
|
Post subject: RE: writeback journal only with telling the kernel to do so
Posted: Nov 01, 2008 - 10:06 PM
|
|
Joined: Oct 03, 2008
Posts: 20
Status: Offline
|
|
| FWIW - dir_index is now enabled by default on most distros, so it shouldn't be needed any more. |
|
|
| |
|
|
|
|
|
|
|
Post subject: RE: writeback journal only with telling the kernel to do so
Posted: Nov 27, 2008 - 06:38 AM
|
|
Joined: Dec 01, 2006
Posts: 751
Status: Offline
|
|
snvv, with tune2fs you can add any of the mount options supported by ext3. Three of the many options are journal_data, journal_data_ordered, and journal_data_writeback.
It would be foolish to add more than one of them as I'm not aware of which option takes higher precedence if multiple conflicting options are given.
GullibleJones - it is, but I would keep it for reference since it _is_ a performance factor that can be set or cleared. |
_________________
"Cool was never cool until the cool guys at Cool industries developed a cool new product: Cool."
|
| |
|
|
|
|
|
|
