my understanding is that we have several issues at hand 1. use of ssds etc instead of hds. here optimising for rotational delays etc is not so relevant. ssds etc are more and more important these days, while hds are still important. 2. unbalanced devices occur frequently in raids. Sometimes it is bad to just use the simple strategy of using first available device, such as in disk oriented md raid10,far, where striping is key to its perfomance. In many cases the simple strategy is probably the best. Where the balance is very uneven, eg. using ssd and hd together, I wonder if you should have some caching strategy, ie caching raid, we already do this with kernel space buffers. It could be a general layer on top of md raid. some technology uses a caching approach having tape as the slowest media. 3. migrating from one raid type to another, some of these are more important than others. IMHO migrating from current md raid10 RAID-0+1 to RAID-1+0 layouts would be quite important. keld
