NetBSD-Users archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: raidframe and disk caching



Pierre-Philipp Braun <pbraun%nethence.com@localhost> writes:

> here's a little but not necessarily simple question.  It's noted in
> the official RAIDframe guide that one should disable disk _write_
> caching if there is no Uninterruptible Power Supply installation.

I find it surprising that this is in the raidrame guide.  There are two
issues:

  write caching on disks is unsafe, unless there is a guarantee that
  anything acknowledged by the disk will make it to stable storage even
  if there is a power failure.  Typically this involves UPS or
  battery-backed ram cache on the disk controller (in high-end hardware
  raid controllers).

  RAID5 in software is scary, because it involves read-modify-write
  cycles of stripes that involve logical blocks that are not being
  written.  Or perhaps RAID5 without battery-backed caches is scary.  I
  have dealt with this by only using RAID1.

The first issue about has nothing to do with raidframe.  With RAID1, I'm
not aware of extra considerations for raidframe RAIS1 vs just using a
disk.

> Ok.  Now what happens if there is a kernel panic? (Yep, this can
> happen.)  Technically, wouldn't it be the same issue?

A panic should result in long enough time between the last write to the
disk (sent to the controller) and the reset from the reboot for the data
to all be written.  In theory writes are ordered such that corruption
will not occur, even though individual files in the process of being
written may have incorrect data blocks.

Attachment: pgpkxlCHRwk4H.pgp
Description: PGP signature



Home | Main Index | Thread Index | Old Index