Re: Question: why does dbench take so much longer to run on XFS then ext2 file system

[Stephen Lord <lord@xxxxxxx>]

Philip Chiang wrote:

> Hi, I am trying to make comparisons between ext2 and xfs file system. I ran dbench on my 
> system with 150 clients on the ext2 file system and on the xfs file system. I ran this 
> test several times, but the results are similar. On xfs, dbench takes 211 mins to finish. 
> Where on ext2, it takes 60 mins. I want to ask "why it has such a big difference on 
> the two file system?" One more note, when dbench is running on xfs, the whole system 
> tag downs, CPU usage is at 99% all the time, is this normal?


Such a discrepency does not seem right, although at dbench 150 on a 128M 
box you are severely
overdriving the memory. Anyone who runs hardware at this load level in a 
production environment
is not going to get much out of it. It's a bit like running a news feed 
on your palm pilot. If you are
trying to compare performance then compare loads which are realistic.

I am doing some comparison runs here, but I have a  question and a comment.

o first, which xfs options do you have turned on? Did you enable 
extended attributes and dmapi?
   if you did then you really need to turn it off in a comparison - or 
get the ext2 equivalent code
   going. This would not however account for such a large difference.

o A more apples with apples comparison would be ext3 (without data 
journalling) vs xfs.
   XFS is going to add the cost of journalling to operations, ext2 is not.

Steve