Dave Wreski wrote:
> Hi all. I've read the most recent HOWTO, and still have a few questions. I plan
> to implement RAID-1 on a 586 with two 6G IDE disks. Most files are greater than
> 4M, if that makes a difference.
I don't think so.
> If I want to experiement with different chunk sizes, must I rebuild the array
> for each different test? Will that destroy the existing data?
As far as I know, you have to rebuild in order to change the chunk size (= loss of
data).
> Must I make a
> new filesystem after rebuilding the array each time?
With raidtools 0.90 you do not have to make a mke2fs on the array (I think it is
mentioned in the doc).
> Surely I could test this to find out, but my test machine is a 486 with 32M, and
> it takes quite a long time to build the array. Is it worth experimenting with
> chunk sizes on a mirror? What factors need to be taken into account when
> changing the sizes?
Changing the chunk size on a stripe (maybe it is similar to a mirror) can increase
performance. I always took sizes like 4k, 8k, 16k, 32k and so on for tests.
> How can I break the mirror? I see that by stoping the raid subsystem, I can
> mount the individual disk. Is there anything further that needs to be done?
In order to use each single disk you have to delete the entries in /etc/mdtab (I
think you are using mdutils 0.36).
> Is there an upgrade path from the 0.36 raid to the current 0.90 code? I recall
> there being some problems at one point with compatibility?
There should be an upgrade option without destroying data - but I don't have
experiences with it (I always used the 0.90-tools at initialization time of the
arrays).
Hope this will help you, Dietmar
