Re: XFS performance

["Alexey Tsyban" <leshats@od.anything3d.com>]

> Alexey Tsyban wrote:
>
>>
>>Hi,
>>
>>I just have tested performance of xfs and ext3 on my new Dell PowerEdge
>>1550 RERC3/DLC RAID with 3 36Gb disks in RAID5, Dual PIII 1000. Running
>>linux slackware 8.0 glibc 2.2.5 with linux 2.4.17 from cvs tree.
>>Mkfs options -l size=8192b, partition size 30Gb
>>mount options logbufs=4,osyncisdsync
>>
> Numbers removed
>
>>
>>As I see ext3 is match faster. Why??
>>May be I should use another mount options or mkfs options to have
>>better performance?
>>
>
> With software raid5, xfs does not like having its log in the middle of
> the drive,
> the raid5 code and xfs log writes do not interact well, but that should
>  not be an issue
> here.
>
> Some of those numbers my puny little scsi drive beats quite nicely, and
> on others I ran out of cpu power (2 x 450MHz PIII with 128M).
>
> You say you have 3 drives configured as raid5 - how? What is the stripe
> unit? I do not know for certain, but 3 seems like a funny number. Can
> you run xfs_growfs -n /xxx where /xxx is the mount point? It is
> possible to teach xfs about stripe units of raid devices, but getting
> it right can be tricky.
>
raid5 is maked by PERC3/DCL (AMI Megaraid) setup utility. I dont know what
is spripe unit.
Yes, 5 disks is better than 3 but in this type of server there is only 4
disk holes and I have only 3 disks.

xfs_growfs -n /.0
meta-data=/.0                    isize=256    agcount=30, agsize=262144
blks
data     =                       bsize=4096   blocks=7681070, imaxpct=25
         =                       sunit=0      swidth=0 blks, unwritten=0
naming   =version 2              bsize=4096
log      =internal               bsize=4096   blocks=8192
realtime =none                   extsz=65536  blocks=0, rtextents=0

> Are you running ext3 with data journalling or without?
>
ext3 has a default configuration, so only metadata is jornaling I think.


Alex
-- 
Alexey Tsyban
Realtor3D Corp. System Administrator
ICQ 10195188