NetBSD-Users archive

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

Re: NetBSD Raid5, slow write speeds, using big disks?!



Thanks Dave, i will most likely have a look at Raid10, which according to wikipedia is two raid1 and one raid0:
raid1a - disk1,disk2
raid1b - disk3,disk4
raid0 - raid1a,raid1b

And might look into raid5 with 3 disks as Greg suggested.
Funny thing is, that i can swear i looked for like best raid config for 4 disks 6 months ago and raid5 was suggested... :D
Would be nice if wikipedia had some rough estimates of speed, like your single drive has X as write speed. Raid0 will have X write speed, raid5 X/4, raid10 X/2 or something like that...

Br Nicklas


On 6/22/26 9:05 PM, Dave B wrote:
If you decide to stay away from ZFS, it still might be best, as I
understand it, to avoid RAID 5 specifically, unless your system
has a hedge against the "RAID 5 write hole".  (Acknowledgment: not
an original opinion on my part; rather, advice which, over the
years, already gets posted to this list from time to time, IIRC).

I'm not claiming to be a RAID or RAIDframe expert, so I won't try
to explain the "write hole"; but suffice it to say that after
experiencing enough problems with RAID 5 (on less-than-rock-solid
hardware & in potentially flakey power-environments), my life was
improved by deciding to stop using it...  This was on big "-ish"
disks at the time, though probably an order of magnitude smaller
than your current ones.

I believe RAID 1, RAID 10 & RAID 6 either don't suffer from
"write hole" issues or at least are less susceptible to them.
[I'd recommend confirming this via more reliable sources; not
stated here as fact].  So RAID 1 is what I now stick with on
RAIDframe generally, and I also use hardware RAID 10 & RAID 6 on
systems with applicable controllers.

Building on previous remarks that a 4-drive set-up isn't optimal
for RAID 5, it's worth adding that 4-drive set-ups *are* nicely
conducive to RAID 10 & RAID 6, and more feasible to align with
stripe & buffer size; though wouldn't include any hot-spares.

The "cons" of RAID 10 & RAID 6 (compared to RAID 5) are that they

  - will only offer you ~14TB total storage (rather than your 4-
    drive RAID 5's ~21TB)
  - and AFAIK aren't directly supported by RAIDframe
    - however, RAID 10 can be achieved by raidframe-on-raidframe,
      at the slightly cumbersome cost of setting up & managing
      three raidN devs (a RAID 1 of two RAID 0's -- or maybe a
      RAID 0 of two RAID 1's, I forget which is RAID 10 & which is
      RAID 01...)

The "pros" of RAID 10 are that it

  - will still survive the loss of 1 drive, just like RAID 5; and
    in uncommon cases, even that of 2 drives.  I wouldn't bank on
    the latter, however
  - should be *really fast* for both reads & writes (comparatively
    speaking)

The "pro" of RAID 6 is that it will survive loss of 2 drives,
provided you can get it working with RAIDframe... although I
believe it would be comparatively slow.

The above is all stated from perceptions & recollections, my
apologies if I have any errors in, e.g., objective facts.  Again,
best to confirm anything critical your data might depend on from
more official sources.

Best, -D




Home | Main Index | Thread Index | Old Index