On Thu, May 30, 2019 at 07:55:45PM +0200, keld@xxxxxxxxxx wrote: > On Thu, May 30, 2019 at 07:50:45PM +0200, Reindl Harald wrote: > > > > > > Am 30.05.19 um 19:12 schrieb keld@xxxxxxxxxx: > > > On Thu, May 30, 2019 at 06:01:29PM +0200, Reindl Harald wrote: > > >> > > >> > > >> Am 30.05.19 um 17:58 schrieb keld@xxxxxxxxxx: > > >>> so you will not get performance gains if you read one file sequentially in raid10,near > > >>> nor md raid1, reading 2 different files concurrently theoretcally should give the same > > >>> speed in md raid1 an mdraid10,near - but I think raid10,near only reads from one device. > > >>> so it is a driver issue. > > >> > > >> but why? > > >> > > >> you have two simple mirrors, read from both disks, one half from mirror1 > > >> and the second from mirror2 > > > > > > if you have one file and you read every other block from each mirror then you loose > > > a lot of reading capacity. > > > > > > say you read block 0 and 2 from mirror1 and block 1 and 3 from mirror2, then > > > after reading block 0 from mirror1 you have to skip block 1 on mirror1 to read block 2, > > > given that the disk is rotating anyway to advance to block 2. you could just as well read block 1 from mirror1 > > > so all your reading frem mirror2 is unnessecary, and you could use mirror2 for other things. > > > > > > it is a matter o the disk drivers strategy. > > > > on a SSD RAID? seriously? > > no, only on hard disks, sorry for not pointing that out the drivers are quite old and mostly written for hard drives. ssd drives are handled by the same code. keld
