Todd Denniston wrote:
Joe Tseng wrote, On 03/26/2008 10:37 AM:
I saw a few people respond with saying how hardware RAID is overkill
for home use. I had the system drive in my RH9 RAID1 file server at
home die on me last year; although I got a new drive and FC6
recognised the RAID immediately I'm not sure whether my recovery was
due to software resilency or dumb luck. I'm currently working on
gathering parts for a RAID5 file server as a replacement.
1) If a RAIDed drive dies in a soft RAID setup can I assume I can't do
a hotswap?
based on my limited information, I would say soft RAID only has a
_chance_ of doing hotswap if the hardware connected to it can do
hotswap. however I don't know if md supports "hotswap" or even being
told "stop using the drive temporarily so I can turn it off in hardware
and swap it".
2) If my system drive dies again would a new system recognize my RAID5
array?
That is the beauty of soft raid under Linux, as long as the RAID5
routines remain the same the new kernel can use them.
3) Does soft RAID5 compare favorably against hware RAID5?
Remember, at it's heart all RAID 5 is software.
in general:
hardware raid = controller card/box with a built in CPU doing all the
RAID calculations and spreading the data across buses and drives.
software raid = main CPU of the system doing all the RAID calculations
and spreading the data across buses and drives.
Difference = your main CPU can be used for other things if you have
hardware to offload the job to.
however seeing as most "hardware RAID" uses a very small/slow CPU, your
main CPU may not notice the extra load of software RAID. :)
On top of that on a lot of applications, things are idle (ie IOWAIT) while
waiting for the HW raid to finish, so nothing useful would be done, so it often
makes no difference if the kernel is doing the raid5 work or the hardware raid
is doing the raid5 work.
In the HPC market this is almost always true, as you don't generally overload
the machine with more processes that cores, and the applications are generally
not setup to do async io.
Roger
