On 03/05/20 06:44, Caveman Al Toraboran wrote: > hi - i'm to setup my 1st RAID, and i'd appreciate > if any of you volunteers some time to share your > valuable experience on this subject. > > my scenario > ----------- > > 0. i don't boot from the RAID. > > 1. read is as important as write. i don't > have any application-specific scenario that > makes me somehow favor one over another. > so RAIDs that speed up the read (or write) > while significantly harming the write (or > read) is not welcome. > > 2. replacing failed disks may take a week or > two. so, i guess that i may have several > disks fail one after another in the 1-2 > weeks (specially if they were bought > about the same time). > > 3. i would like to be able to grow the RAID's > total space (as needed), and increase its > reliability (i.e. duplicates/partities) as > needed. > > e.g. suppose that i got a 2TB RAID that > tolerates 1 disk failure. i'd like to, at > some point, to have the following options: > > * only increase the total space (e.g. > make it 3TB), without increasing > failure toleration (so 2 disk failure > would result in data loss). > > * or, only increase the failure tolerance > (e.g. such that 2 disks failure would > not lead to data loss), without > increasing the total space (e.g. space > remains 2TB). > > * or, increase, both, the space and the > failure tolerance at the same time. > > 4. only interested in software RAID. > > my thought > ---------- > > i think these are not suitable: > > * RAID 0: fails to satisfy point (3). > > * RAID 1: fails to satisfy points (1) and (3). > > * RAIDs 4 to 6: fails to satisfy point (3) > since they are stuck with a fixed tolerance > towards failing disks (i.e. RAIDs 4 and 5 > tolerate only 1 disk failure, and RAID 6 > tolerates only 2). > > > this leaves me with RAID 10, with the "far" > layout. e.g. --layout=n2 would tolerate the > failure of two disks, --layout=n3 three, etc. or > is it? (i'm not sure). > > my questions > ------------ > > Q1: which RAID setup would you recommend?
I'd recommend having a spare in the array. That way, a single failure would not affect redundancy at all. You can then replace the spare at your leisure. If you want to grow the array, I'd also suggest "raid 5 + spare". That's probably better than 6 for writing. but 6 is better than 5 for redundancy. Look at having a journal - that could speed up write speed for raid 6. > > Q2: how would the total number of disks in a > RAID10 setup affect the tolerance towards > the failing disks? > Sadly, it doesn't. If you have two copies, losing two disks COULD take out your raid. > if the total number of disks is even, then > it is easy to see how this is equivalent > to the classical RAID 1+0 as shown in > md(4), where any disk failure is tolerated > for as long as each RAID1 group has 1 disk > failure only. That's a gamble ... > > so, we get the following combinations of > disk failures that, if happen, we won't > lose any data: > > RAID0 > ------^------ > RAID1 RAID1 > --^-- --^-- > F . . . < cases with > . F . . < single disk > . . F . < failures > . . . F < > > F . . F < cases with > . F F . < two disk > . F . F < failures > F . F . < > . F F . < > > this gives us 4+5=9 possible disk failure > scenarious where we can survive it without > any data loss. > > but, when the number of disks is odd, then > written bytes and their duplicates will > start wrap around, and it is difficult for > me to intuitively see how would this > affect the total number of scenarious > where i will survive a disk failure. > > Q3: what are the future growth/shrinkage > options for a RAID10 setup? e.g. with > respect to these: > > 1. read/write speed. iirc far is good for speed. > 2. tolerance guarantee towards failing > disks. Guarantees? If you have two mirrors. the guarantee is just ONE disk. Yes you can gamble on losing more. > 3. total available space. iirc you can NOT grow the far layout. > > rgrds, > cm. > You have looked at the wiki - yes I know I push it regularly :-) https://raid.wiki.kernel.org/index.php/Linux_Raid Cheers, Wol
