On Tue, Jun 11, 2019 at 7:40 AM <keld@xxxxxxxxxx> wrote: > > On Tue, Jun 11, 2019 at 04:11:52PM +0800, Guoqing Jiang wrote: > > Hi, > > > > On 6/11/19 3:43 PM, keld@xxxxxxxxxx wrote: > > >thanks for this patch > > > > > >I think we should change the hd algorithm to chose the highest block > > >number at least for the > > >far layout. ther outer blocks have the fastest transfer rates and also the > > >shortest > > >distance for head movement. > > > > I didn't investigate the performance of far layout a lot, seems there > > was one patch > > (commit 8ed3a19563b6c " md: don't attempt read-balancing for raid10 > > 'far' layouts") > > which was aimed to do it, and you were the author, no? ;-). Or I missed > > something. > > yes , I was the author of that patch. > and it solved the problem: to get the drives to stripe, evne if the hd drives > have different transfer rates and rotation speeds. > > what I think I got wrong was that it was using the inner parts of the disks > instead of the outer parts, where the transfer rate is higer and head movement less. IIUC, the patch prefers using smaller LBA ranges, no? And smaller LBA means outer sectors, no? Thanks, Song > > I am suggesting now to reverse this. > > keld
