Performance comparisons for various Linux file systems

Turn from: http://hi.baidu.com/xuzhi1977/blog/item/c5869758dfafbade9d82040a.html

North Asia Data Recovery Center (http://www.datahf.net) Zhang Yu slightly finishing.

The following is the original text:

The following article is my own translation, followed by the original English. The improper place, please advise.

It shouldn't be too new, but I just saw it in the morning of 2006-07-12. Hey........

2006-07-12 15:00 Translation completed

--------------------------------------------------------------------------------------------------------------

Intestines are regret green, yesterday just added the hard disk above the file system or I made into a ext3. If I could build a day to notice this article ...

--------------------------------------------------------------------------------------------------------------- -

Debian Administration

System administration Tips and resources

Many Linux filesystems comparisons are available now, but most of them are old, based on human tasks or older. This benchmark is based on the older generation of 11 hardware (Pentium Ii/iii,eide hard drives) for a single file server.

From initial preparation to publication, the article has produced many changes, suggestions and suggestions for improvement. I'm trying to do some new experiments at the moment. (Answer questions in the original range).

The results will be available in approximately (May 8, 2006)

Hans

Why do benchmark tests?

I found that the quantitative and reproductible benchmark benchmarks use 2.6.x kernel.

Benoit achieved 12 trials in 2003 using large files (1 + GB) on PIII 500 servers with MB RAM. The experiment is very informative, but the results are starting with the 2.6.x kernel, mainly for the base, which specializes in large files (for example, multimedia, science, databases).

Piszcz implemented 21 tasks in 2006 (with 768 MB RAM and a 400gbeide-133 hard drive to simulate multiple file operations in PIII-500). So far, this test seems to be the most comprehensive work on the 2.6.x kernel. However, many of the tasks are man-made (for example, copying and deleting 10,000 empty directories, creating 10,000 new files, recursively splitting files), and applying these conclusions to the real world may be meaningless.

Thus, the goal of the benchmark tested here is to validate some Piszcz (2006) findings, which are found in the Small Business file server (see task Description), specifically for the real world.

Test Basics

* Hardware Processor:intel Celeron 533

* RAM:512MB RAM PC100

* Motherboard:asus P2B

* Hard drive:wd Caviar SE 160GB (EIDE, 7200 RPM, 8MB Cache)

* controller:ata/133 PCI (Silicon Image)

* OS Debian Etch (kernel 2.6.15), distribution upgraded. April 18, 2006

* All optional daemons killed (cron,ssh,samba,etc.)

* filesystems Ext3 (e2fsprogs 1.38)

* ReiserFS (Reiserfsprogs 1.3.6.19)

* JFS (Jfsutils 1.1.8)

* XFS (Xfsprogs 2.7.14)

Selected Test Task description

* Copy from 2nd disk to this test disk in a large file (ISO image file, MB)

* Then copy this ISO again from another location

* Delete two copies of this ISO

* Operation of a file tree (7500 files, 900 directories, 1.9GB), copied from the 2nd disk to this test disk

* Then copy the file tree again from another location

* Delete Two copies of this file tree

* Recursive method to traverse the entire contents of the file tree directory and file tree, copied to this test disk

* Match wildcard characters, find specific files in the file tree

* Use (MKFS) to establish the filesystem (all FS Use default values)

*mount filesystem

*umount filesystem

The above 11 tasks (from establishing filesystem to umounting filesystem) are written in the order of Bash. Run 3 times (report average score). Each order takes approximately 7 minutes, the time to complete the task is in seconds, and the GNU Times utility (version 1.7) records the CPU utilization percentage of the task.