On Fri, 29 Apr 2016, Karel Gardas <[email protected]> wrote: > On Fri, Apr 29, 2016 at 2:38 PM, Kamil Cholewiński <[email protected]> wrote: >> Silly, tangentially related question, perhaps someone knows an answer: >> >> Is there a considerable performance impact to be expected when using an >> odd number of disks in RAID[56] setups? >> >> I mean, e.g. with RAID5, one disk stores parity data, so in a 3-disk >> setup, a 512-byte data block is split between two devices. In a 5-disk >> setup, or in a 6-disk RAID6 setup, similarly the data chunk is split >> between four physical devices, so 512/4=nice number. What about >> situations where 512/3, 512/5, etc? >> >> Am I making sense or garbage? > > If you look into softraid_raid5.c you will see that actual write I/O > is done to only 2 drives: one data and one parity chunk. The thing is > that both data and parity chunks change based on the disk block > position and data length. I really recommend to see the code as it is > nicely commented. Look for sr_raid5_rw and sr_raid5_write. In case of > RAID6 this is more complicated, but write I/O goes to 3 drives (1 data > + 2 parities) depending again on disk block position and data length.
That's brilliant. I'll definitely have a look. Thanks!
