Re: bonnie++: disastrous RAID 1/5 results
martin f krafft wrote:
> I set up a Software RAID System previously. It contains four 120Gb
> harddrives, 8 partitions on each, with the forth drive being used as
> a spare:
[SNIP]
> As you can see, the first is a simple RAID 1, all the others are
> RAID 5. All have a chunk size of 32 and an ext3 filesystem (-b 4096
> -R stride=8). Only md5 has a chunks size of 4 and an ext3 filesystem
> with blocksize 1k and stride=4 (because it is so small).
>
> Now, if I use bonnie++ on a dual 2GHz with 2Gb RAM on a reiserfs
> partition on a 120Gb Maxtor IDE harddrive, I get the following
> results:
[SNIP]
> If I use the same options on aforementioned RAID system (1GHz single
> processor machine, 1Gb RAM and ext3), I get these results:
[SNIP]
> Eyeballing, there is a factor 4-8 of a difference. This can't be
> mere processor power, can it? Does this mean that my RAID is
> performing worse than crap?
>
> Does anyone have any clues, any suggestions for a better test, or
> any improvements for my RAID system? Russell?
The first thing I notice is that you compare two different filesystems:
reiserfs and ext3. It is a known fact that reiserfs is in most workloads
*much* faster than ext3.
Redo your test with either ext3 on the 2GHz machine or reiserfs on the
RAID5/1GHz machine.
Next:
How are your three (four) disks connected? Do they all have a seperate
IDE channel or do some of them share one?
Do all disks have DMA turned on?
Compare the output of 'hdparm -vi /dev/hdX' from the 2GHz machine with
the values from the 1GHz machine. Make them the same if they aren't.
Next:
Try with the defaults values for chunk size, algorithm and ext3 stride.
How does it look now?
Next:
Use an external journal for ext3 that is located on a RAID1.
Cheers,
Juri
Reply to: