[geeks] RAID levels

[Gregory Leblanc]

On Sun, 2001-11-18 at 20:41, Joshua D Boyd wrote:
> On Sun, Nov 18, 2001 at 02:54:08PM -0600, Bill Bradford wrote:
> > So, I've got an A1000 full of 9G drives on the way to attach to
> > the SunHELP server (sunfreeware archives, BIG BAD LIST SEARCH ENGINE,
> > here we come!).  Suggestions on RAID levels?  I'l be using the latest
> > version of SDS (4.2.1?).  Content will be mostly reads (~90%) with 
> > the only writes being manual updates to web content, MySQL database for
> > banner ads, and apache access logs...
> > 
> > so, 0+1 or 5?  Doing mirrored pairs of disks would be a waste of
> > space..
> 
> Err, isn't mirrored pairs pretty much the same disk usagewise as 0+1?

As mirrored pairs, yes.  It's just letting the administrator do the
striping instead of the RAID software

> I don't have specific numbers, but raid 3, 4, or 5 are worst at write time.  
> For reads (under linux at least), the performance is about the same as it is 
> for raid 0.  So, it shouldn't kill your CPU.  Writes, of course, suck the big 
> one, but you probably have more than enough CPU power to spare for now.

Sounds about right.  The more spindles that you have in a RAID 3/4/5
config, the closer read performance is to RAID 0 read speed.  The write
performance on 3/4/5 is mostly due to high latency XOR calculations,
although in 3/4 some of it can be caused by having parity on a disk that
isn't significantly fast.  Anybody know how good the US-II is at XORs?
	Greg