On Mon, 18 Aug 2014 18:32:57 +0000 Markus Stockhausen <stockhausen@xxxxxxxxxxx> wrote: > Hi, > > I'm wondering if there are any pros/cons for handling stripes in memory > exactly in the same layout as on disks. There are several places that could > be simplified if parity disks would be always at the end (DDD..DDPQ). Of > course one would have to do the mapping during read/write operations. > > Cases for optimization would be: > > - replace conditions "i == pd_idx || i == qd_qdx" with "i>=pd_idx" > - run data disk loops starting/ending with pd_idx (triple parity raid someday?) > - remove syndrome disk order calculation functions > - Shaohua's last patch could cross chunks > - ... > > Just in case I missed a discussion I'm interested if patches are welcome. My guess is that you would just end up moving the complexity from one place to another. If I'm wrong and the change ends up making a noticeable improvement in code readability, then I would certainly consider it. But I'm pretty sure it would be a big change, so it would need to bring a real improvement. So I'm sceptical but happy to be proven wrong. NeilBrown > > Markus
Attachment:
signature.asc
Description: PGP signature
