I was wondering if there is support for a raid level where one can 
choose the redundancy level.  Raid 5 allows for one redundant drive, 
Raid 6 for two, but there's no raid level that would support arbitrary 
redundancy (even to the point of allowing one to add extra redundancy 
later on, by simply putting spare drives to immediate use).

I'm just a casual Software Raid user, and have been using Raid 6 for a 
small 8 TB array for a few years now and I realize that there's quite 
some CPU overhead involved for such redundancy, but for my purposes 
speed is not a relevant concern (10 MB/sec is enough).

--John


--

[ message continues ]

As far as i know, theres not support for this, however, i do believe
raid6 uses reed-solomon.


--

[ message continues ]

symform, a startup create 32 parity stipes of 64 original stripes, and
distribute the fragments in different geographies.

they call it, RAID-96 And apparently reed-solomon coded.

http://www.symform.com/features-benefits.aspx

thanks

--

[ message continues ]

I have been toying with the idea of an open source project to achieve a 
more generalized functional spec. than what symform has created. I was 
thinking of calling it RAINcloud (redundant array of independent nodes)

I started drawing up the requirements over the holiday weekend.


-- 
Andrew Dunn
http://agdunn.net

--

[ message continues ]

Would that be nodes as in different systems? In that case you should
check glusterfs and provide a raincloud module for it.

On the other hand if nodes is just local disks glusterfs might still
be a good way to implement and test your algorithm before putting it
into the kernel.

MfG
        Goswin

PS: Isn't the Reed Solomon for x+1 disks == XOR and x+2 disks p+q
parity? I.e. raid4/5/6 are Reed Solomon are just the first 2 cases for
Reed Solomon.
--

[ message continues ]

Actually you can choose the number of copies in raid10, via the -n
option. 

Best regards
keld
--

[ message continues ]