ext3 performance with hardware RAID5

[Simon Oliver]

> You didn't specify the size of the external journal device.  
> In our testing we found that having a larger journal 
> (internal) made a large difference in performance.

>> For these tests I created a 256MB partition in the mirrored disk set for
the log device: /dev/sdb6.

The internal log was the default size (32MB?). The external journal device
is 256MB.

>From what I've read the journal size doesn't come into play until the load
becomes very high such as with lots or NFS IO.  

I've also read that "if the journal ever gets full, the throughput pauses
for a while until every transaction in the journal has been checkpointed
onto the main device.  One way to avoid this is to reduce the flush time by
altering /proc/sys/vm/bdflush.  This way transactions get flushed to disc
before the journal fills."

Do you have any experience with this?  Is there any way to monitor how full
the journal gets?

> I believe 
> that the external journal will be the full size of the device 
> (could be wrong of course) so it is probably worthwhile to 
> ensure that the external journal is the same size as the 
> internal journal (can be specified with "-J size=<mb>" for an 
> internal journal, and "mke2fs -s 4096 -O journal_dev 
> /dev/sdb6 <blocks>" for the external journal.

Good point - I'll make that change and rerun the tests.

-- 
  Simon Oliver