Re: RAID 0 minixfs?

[hyc@locus.com (Howard Chu)]

  Now, say I've got two disks that are exactly the same size (which, I
  do).  Has anyone hacked on MiNT's minixfs to implement RAID 0
  ("striping") over two or more disks?

  I think this would be pretty neat, even if our halt-the-system I/O got
  in the way a little.

  How would you do it?  I think a lot of RAID systems use 8 or more disks
  and split the bits apart(!), though they tend to do this in hardware.  I
  was thinking whole bytes, or maybe even longwords...

  Would this actually speed things up at all?  Or would the halting I/O
  stall things too much?

No, it sounds like there's potential to speed things up quite a bit. Just
for simplicity's sake, I would split things apart block-wise. I.e., every
512 byte sector goes to another device. This approach makes bookkeeping
very simple, and eliminates the requirement of having exactly identical
drives. (You just use as many blocks as the smallest drive in the chain.)
The rationale is that any SCSI disk can buffer some number of blocks at
full speed, so if you're writing (for example) 1024 bytes to 2 drives, you
effectively get 10 MB/sec performance on that write. (or however fast your
bus will go...)
 -- Howard