[NMLUG] Hardware RAID5 vs Software RAID5

[wjl at icecavern.net (Wesley J. Landaker)]

On Thursday 09 November 2006 11:03, Tim Emerick wrote:
> One of the reasons the author decided to use a hardware raid5 was that in
> the event of failure the ability to rebuild is in the hardware bios.  It is
> basically OS independant.  He also notes that it appeared difficult to
> recover from a failure using linux software raid and couldn't easily figure
> out how to rebuild the raid in event of failure.

Keep in mind that most low-end and mid-range "hardware" RAID controllers that 
are not true hardware RAIDs. Usually they just provide a XOR and/or DMA 
engine, maybe some buffering, but has it all to be driven by a software 
driver. What this means is that you *may* get better performance if the 
driver uses the hardware RAID acceleration functions properly and 
efficiently, but it's not necessary OS independent, as many of these cards 
only have Windows drivers, and don't have open specs. Sometimes, even if they 
have Linux RAID drivers, they may be slower in "hardware" RAID mode than just 
using them as normal non-RAID controllers and doing software RAID with the 
normal Linux dmraid methodology. 

Also, most hardware RAID solutions are tied to the controller--you can't span 
a RAID across multiple controllers, and you can't always use a drive set 
built on one controller on a different vendor's controller.

Anyway, be very careful before assuming that a "hardware" RAID solution is 
going to be the best bet. I've found that Linux software RAID generally gives 
the best performance and flexibility compared to almost any "hardware" 
solution, until you start looking into the really high-end setups (and even 
then, you often sacrifice flexibility).