On Fri, 2013-07-05 at 09:36 -0400, Phil Turmel wrote: > You picked "redundancy". Leaves only only one axis to consider: speed > vs. capacity. Well thinking about that "raid6check" tool you told me over in the other thread,... which AFAIU does what I was talking about, namely telling me which block is the correct one if I have bad blocks (and the disk itself can't tell) and not whole drive failures,.. where at least a two-block copy RAID10 would not be able to... ...then I think RAID6 is THE solution for me, given resilience has the highest priority, as RAID10 with c/f/o=3 cannot do that. > Precisely. This is "raid10,near3". You can look up the "offset" and > "far" variants. Actually I had written some ASCII art visualisations for the Debian mdadm FAQ (and perhaps also the md(4) manpage) yesterday... I just wait for some minor answers from Neil to publish them... so I already had a look at these. But AFAIU, they make absolutely no difference for resilience... and while far/offset improve performance in some use cases,... they make it worse in others. > > And that gives me again, any 2 disks... but so much better performance? > Dramatically. I guess I'll do some benchmarking.. ;) > At four drives, no. Any two. With five, there are some combinations of > three missing drives that'll still run. Was thinking about that as well... but as you said it's a "it might survive"... not a "it will survive", right?! > > Any ideas about the layout? i.e. left-symmetric-6, right-symmetric-6, > > left-asymmetric-6, right-asymmetric-6, and parity-first-6 ? > Certainly not any of the "-6" suffixes. Ah.... I just see that the RAID5 layouts are also for RAID6 ^^ > Those isolate Q on the last > disk, hurting streaming performance, and setting up the possibility of > uneven performance when degraded. The default left-symmetric gives the > best chunk distribution for general use. I see... I was just searching for but haven't found anything really good... actually nothing at all ^^ Does anyone know some place where the layouts for RAID5/6 are (thoroughly) explained? Cheers & thanks, Chris.
<<attachment: smime.p7s>>
