Michael, it can make a _big_ difference.
Let us take a look at the differences:
- mirrored stripesets:
You bind four disks with 36 GBytes together to form a 144 GByte stripeset. Next you create a second stripeset that way. On top of that you create a mirrorset to keep the contents of both stripesets equal. The resulting mirrorset and what the hosts sees is, of course, 144 GBytes.
What happens when a single physical disk fails? On all implementations that I know this hierarchy is a strictly layered one - the upper layer (your mirrorset) does not know how the lower ones (your stripesets) are implemented. The result is a complete failure of the stripeset.
Now, the HSG does not permit this type of layering, but on host-based solutions I have seen, this results in the defective stripeset being 'split off' from the mirrorset.
You have to create a new stripeset. You cannot simply replace the failed disk with a new one, because this is missing the metadata. You need to create a new set which updates the metadata on all physical disks. Only then you can add the 'repaired' stripeset back to the mirrorset.
Now, the mirroring mechanism does not know anything about the underlying object - it could be a 144 GByte disk as well. It will start mirroring ALL 144 GBytes again, just because a single 36 GByte disk has failed.
- striped mirrorsets:
As you correctly say, this is what the HSG can do. You take two 36 GByte disks and create a mirrorset from them. You do this four times and get four mirrorsets, each with a size of 36 GBytes. Next you go and create a single stripeset that uses those four mirrorsets. That stripeset is, surprise, 144 GBytes in size. So, for the host this makes no difference.
Now, what happens when a single physical disk from a mirrorset fails? Again, the host can continue to work. Even better: the stripeset continues to work, too. As long as there is an appropriate spare-disk the HSG will go and add it to the mirrorset.
And here is the big difference:
You loose a single physical disk with 36 GBytes of data and you only need to copy that 36 GBytes over to the replacement disk.
Even better - if you loose only one disk in both stripesets of the first example you have lost the entire LUN. For the second example you would have to loose both disks of a single mirrorset.
Perhaps someone with a stronger mathematical background than me can calculate the propabilities(sp?) for a total failure for each example, please?
.
