[ale] Comparing EXT4 and JFS

Greg Freemyer greg.freemyer at gmail.com
Wed Feb 4 13:38:41 EST 2009


Jeff,

Ext4 just went "stable" as of 2.6.28. (ie. Christmas time).  Thus I
assume it is just now getting much real world use.

I would be cautious of moving to it this soon for an important server.

As to performance, I don't know much about JFS, but XFS is notoriously
slow at file create / delete.

People often use a simple untar of the kernel source as a file create
benchmark, and the deleting of that as a file delete benchmark.

Seems like XFS used to take a couple minutes to do the delete!!!
Obviously not a good choice of filesystem if you plan to do a lot of
kernel development.

OTOH, xfs had a great reputation for working with large data streams.
(ie. video files, etc.)  So you need to be sure to test with the same
usage pattern you expect from your actual use.

Greg

On Wed, Feb 4, 2009 at 10:37 AM, Jeff Hubbs <hbbs at comcast.net> wrote:
> I'm preparing a big file server and I was trying to compare the
> performance of EXT4 and JFS on a hardware RAID array.  I've got two
> identical arrays, each on their own 3ware 9690 card. All tests were
> performed with the machine doing nothing else (save being shelled into
> from my laptop).
>
> Bonnie++ output is shown below.  I was seriously considering using JFS
> for the big arrays, so I'm concerned that JFS' performance in create and
> delete seemed to be really horrid by comparison.
>
> As you can see, I experimented with using an external log device for JFS
> and with changing from RAID 6 to RAID 5.
>
> Does this track you guys' experiences?
>
> - Jeff
>
> RAID6 arrays on 3ware 9690; 64KB chunk size, 7x146GB 15000RPM drives
>
> /dev/sdb (ext4) RAID 6
> Version 1.93c       ------Sequential Output------ --Sequential Input-
> --Random-
> Concurrency   1     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block--
> --Seeks--
> Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP
> /sec %CP
> jupiter         16G   517  94  6429   1  6889   1  1761  74 306476  30
> 1021  25
> Latency             14989us    3683ms    4274ms     298ms   16045us
> 19601us
> Version 1.93c       ------Sequential Create------ --------Random
> Create--------
> jupiter             -Create-- --Read--- -Delete-- -Create-- --Read---
> -Delete--
>              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
> /sec %CP
>                 16  9005  15 +++++ +++ 10808  14  9438  15 +++++ +++
> 11191  14
> Latency               128us     129us     169us     230us       9us
> 55us
>
> /dev/sdc (jfs) RAID 6
> Version 1.93c       ------Sequential Output------ --Sequential Input-
> --Random-
> Concurrency   1     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block--
> --Seeks--
> Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP
> /sec %CP
> jupiter         16G   997  90  7822   1  7933   1  1462  62 312938  28
> 1652  13
> Latency              7509us    1728ms    1723ms     264ms   27288us
> 138ms
> Version 1.93c       ------Sequential Create------ --------Random
> Create--------
> jupiter             -Create-- --Read--- -Delete-- -Create-- --Read---
> -Delete--
>              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
> /sec %CP
>                 16   934   2 +++++ +++   375   1   261   1 +++++ +++
> 107   0
> Latency               369ms      46us    1837ms    1255ms       5us
> 17067ms
>
> /dev/sdc (jfs) RAID 6; external log device
> Version 1.93c       ------Sequential Output------ --Sequential Input-
> --Random-
> Concurrency   1     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block--
> --Seeks--
> Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP
> /sec %CP
> jupiter         16G   955  86  7488   1  7673   1  1305  57 311921  27
> 1532  10
> Latency             14861us    3328ms    1612ms     328ms   26300us
> 27019us
> Version 1.93c       ------Sequential Create------ --------Random
> Create--------
> jupiter             -Create-- --Read--- -Delete-- -Create-- --Read---
> -Delete--
>              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
> /sec %CP
>                 16  4491  12 +++++ +++  1619   3   582   3 +++++ +++
> 149   0
> Latency               676ms      47us    1102ms     840ms       7us
> 10821ms
>
> /dev/sdc (jfs), RAID 5; no external log device
> Version 1.93c       ------Sequential Output------ --Sequential Input-
> --Random-
> Concurrency   1     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block--
> --Seeks--
> Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP
> /sec %CP
> jupiter         16G  1005  92 13256   1 13236   1  1523  65 334584  28
> 1732  11
> Latency              7781us    1892ms    1091ms     257ms   24041us
> 56333us
> Version 1.93c       ------Sequential Create------ --------Random
> Create--------
> jupiter             -Create-- --Read--- -Delete-- -Create-- --Read---
> -Delete--
>              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
> /sec %CP
>                 16  1009   2 +++++ +++   436   1   291   1 +++++ +++
> 116   0
> Latency               305ms      48us    1548ms    1043ms       7us
> 16346ms
>
>
> _______________________________________________
> Ale mailing list
> Ale at ale.org
> http://mail.ale.org/mailman/listinfo/ale
>



-- 
Greg Freemyer
Litigation Triage Solutions Specialist
http://www.linkedin.com/in/gregfreemyer
First 99 Days Litigation White Paper -
http://www.norcrossgroup.com/forms/whitepapers/99%20Days%20whitepaper.pdf

The Norcross Group
The Intersection of Evidence & Technology
http://www.norcrossgroup.com


More information about the Ale mailing list