特殊:Badtitle/NS100:UbuntuLTSP/FilesystemOptimize

来自Ubuntu中文
Wikibot留言 | 贡献2010年5月20日 (四) 00:29的版本
(差异) ←上一版本 | 最后版本 (差异) | 下一版本→ (差异)
跳到导航跳到搜索

{{#ifexist: :UbuntuLTSP/FilesystemOptimize/zh | | {{#ifexist: UbuntuLTSP/FilesystemOptimize/zh | | {{#ifeq: {{#titleparts:UbuntuLTSP/FilesystemOptimize|1|-1|}} | zh | | }} }} }} {{#ifeq: {{#titleparts:UbuntuLTSP/FilesystemOptimize|1|-1|}} | zh | | }}

This page is specific to Ubuntu versions 8.04, 8.10, 9.10

If you find this information applicable to additional versions/releases, please edit this page and modify this header to reflect that. Please also include any necessary modifications for this information to apply to the additional versions.


It is a good idea to optimize the filesystem for use by many simultaneous users. This page will try to give you some hints as to what you can do to optimize filesystem I/O using common practices.

ext3

Using the 'noatime' flag will greatly help in I/O. Simply edit your /etc/fstab file and append 'noatime' to the end of the filesystem options, and remount. Example /etc/fstab:

UUID=e84ae282-d80b-4db4-8b0a-6fx0314508eb /               ext3    errors=remount-ro,acl,noatime 0       1

Notice the presence of noatime which is appended after acl. Now all we have to do is remount our filesystem (No need to reboot):

sudo mount -o remount /

Repeat the remount for all filesystems you append the noatime flag to.

Links

ext3 latency: http://lwn.net/Articles/328363/

IRC conversation regarding disk I/O

Lns Anyone using VMWare guests on Linux hosts (especially ltsp servers) that have tips to reduce disk i/o? It seems to be a *major* bottleneck for me, causing horrible hangs/delays during some operations (especially during the first ~10min startup of a win2k3 guest).

jammcq Lns: yeah, get faster/more disks
jammcq we've got a big setup with a SAN and vmware with an LTSP server as one of the guests, and it's working VERY well

vagrantc Lns: be sure to use noatime

jammcq I typically go with Raid 10, with 12 or more drives
jammcq 'noatime' tells the kernel not to update the 'time last accessed' when a file is read
jammcq and vmware is doing mostly reading

Lns how about converting to ext4? Thoughts on stability in a production env?

jammcq never used ext4, so no opinion

vagrantc Lns: if you use mutt with mbox files, you *might* have issues seeing new mail. if you use popularity-contest, it won't be able to collect as much useful data.
vagrantc Lns: a middle-ground compromise is relatime
vagrantc Lns: but i'd just put your vmware disk images on a partition with noatime, for sure.
vagrantc there's no sane reason to need that for disk images

alkisg Lns: causing horrible hangs/delays during some operations (especially during the first ~10min startup of a win2k3 guest ==> that's usually the windows update service
alkisg Installing a local WSUS server helps a lot