In message <20051126002526.T18082@chylonia.3miasto.net>Wojciech Puchar writes
>>
>> we are looking for a new server, which should have a RAID array (probably 
>> with SATA disks, as the SCSI price tag is a little bit high) and run under 
>> NetBSD.
>>
>
>what RAID type?
>
>for RAID-0,1,10 buying extra hardware it's not worth of. for RAID-5 it 
>may, but probably will exceed cost of extra discs to make it as RAID-1/10

Wojiech, you seem deeply confused as to why people deploy RAID in
server environments (more generally, shared hardware resources). You
seem to taticlty assume the point of RAID in such environments is as a
substitute for backup. In reality, the reasons for RAID are not
primarily to avoid backup; point is to avoid *outage*, as seen by
users of the shared resource.  

[[I could add a long tutorial about how increasing the number of
single-point-of-failure-devices decreases expected MTFB of a system.
I could add more, to the effect that RAID originally stood for
Redundant Arrays of Inexpensive Disks, and how these
days,. "inexpensive" disks means consumer-grade disks which, all too
often, have non-Poisson, clustered, failuress, making RAID a very
useful technique which masks single-device disk failures, thus
preventing them from becoming a system failure. Indeed, RAID-6 is
gaining popularity precisely because a gives a wider window to replace
one failed drive before a second failure leads to data loss, and an
unwanted outage whilst data is recovered from backup.]]

Wojciech, your recent messages, to the effect that RAID is not a
subsitute for backup [and thus that somehow that RAId is not useful;
see [*] below], is what I can only describe as remedial advice. Repeating
that advice seems (to me at least) to assume that subscribers to this
list, who use RAID (for very good and sufficient reasons) are in need
of your remedial advice: in other words, that the subscribers of this
list are bordering on clueless. Which is pretty offensive.

Is that *really* how you intend to come across? 

Let's crunch some numbers. At yesterday's online US prices,
consumer-grade 200GB SATA drives were $200 per drive.  As a strawman,
take another recent poster's 2TB filesystem; that's 5 drives for
data. (Maybe 6, given marketing megabytes).  To mirror each drive will
double your cost, roughly $1,000.  Given the space power and heat
costs (and higher failure rates?)  of 10 or 12 drives, even a
$1000-$1,200 hardware RAID-5/RAID-6 can be an *extremely* cost-effective
tradeoff.



[*]: If anything, you seem to suggest that RAId is not necessary given
a suitable backup strategy. That is only true if you put an ....
*extremely* low cost both on downtime, and on the operator time
required to restore from backup after failure of a disk.  IMO, that's
likely to piss off both those who've suffered lost time during
outages, or lost fresh data which hadn't yet lived through a backup;
and also piss off those who suffer the thankless task of performing
restores from backup media.

The first two or three times I read your first two or three messages
in this thread, I was left with the impression that you're thinking
about two usage scenarios: home or small-office use (where the
operator is perhaps the only user affected by an outage); and
indentured servants .... er, excuse me, graduate-student labour.