🚀 go-pugleaf

Ross Irvine wrote:
> 
> Brian O'Connor (boc@ironbark.bendigo.latrobe.edu.au) wrote:
> :
> : Hello,
> :       we are considering setting up a large linux(or bsd) box, one
> : of the things that is plesent about Digital Unix(and others) is
> : logical volumes, I can concatanate multiple disks into one volume,
> : in addition, RAID would be an advantage. Does linux support RAID, and/or
> : LV?
> 
> Can't help you on spanning volumes but as for RAID, the only good RAID is
> hardware raid, software raid is unreliable, NOT as a fault tolerant as
> hardware raid, and places extra strain on the CPU etc.
> 
> regards.
> 

Just a quick warning and correction. Be careful when making statements
and claims in this area, particularly when using term which have
specific meanings for the high availability/continuous computing
industry.

1) 'reliable' means to be consistent i.e. to give the same results when
given the same inputs (N.B. this is not the same as 'correct', something
can be 'reliable' bot consistently 'incorrect'). Thus it is patently
WRONG to say that software is unreliable. Software is absolutely
reliable, whereas hardware suffers from wear, degradation and the
influence of external effects e.g. radiation and is thus much more
unreliable.

2) to be 'fault tolerant' is to allow for, and to continue to operate
after, a fault in a component, usually by the use of duplication of
elements to remove single points of failure. Assuming the responder
above was writing about using a single hardware controller versus using
a single software program, then it is incorrect or even meaningless to
claim that hardware raid is more fault tolerant than software raid.
Either is a single point of failure and therefore by definition not
'fault tolerant'. The normal approach taken in the industry is to use a
combination of both hardware and software, often to back each other up,
to provide true fault tolerance.

Having said that, it is impropable that the original poster really
requires 'fault tolerance'. From the note I would guess that that their
interest is more in 'data integrity', rather than absolute system
availability, combined with ease of use/management.

I would therefore suggest that the sorts of factors to consider are:

1) cost - (cost of raid hardware plus related changes e.g. need to adopt
SCSI and replace existing drives, versus cost of software plus possible
system enhancements to maintain performance)
N.B. RAID, especially RAID 5, extracts a severe penalty in I/O
performance.

2) ease of use e.g. does either approach allow dynamic changes to the
volume organisation.

3) availability - e.g. what hardware raid controllers are available in
Australia (and supported - if not interested in support then probably
not really serious about availability), and do these controllers work
with the intended system/Operating System.

Dave B.

Maybe speaking for myself, maybe not (with 0.5 probability) but
definitely (with probability 0.01) not speaking for anyone else.