[wellylug] preliminary SATA Linux software RAID0/RAID1 comparison

[Wood Brent]

> 
> > > The total size (-s) should be at least equal to your total ram (if not
> > more)
> 
> Whoops .. must've been half asleep when I was posting that.  The size should
> be at least *twice* the ram in your machine. :)
> 
> > > to ensure that you don't end up testing Linux's disk cache rather than
> the disk read speed.


Yep. I've left it running: 

bonnie++ -d /raid0 -s 2048 
bonnie++ -d /raid1 -s 2048 
bonnie++ -d /home  -s 2048 

look like reasonable parameters. I'll see what it says when I get back home
tonight. Given that this is software RAID, bonnie++'s %cpu load figures will be
interesting for the various runs. /home is non-raid so should show a much lower
cpu use. 

Dunno about RAID0 & RAID1, my guess is that as RAID1 actually does twice the
I/O, it will be significantly slower on writes, (where good hardware RAID can
gain). 

If there is good read balancing, as suggested by the RAID page Ewen mentioned,
then kernel 2.6 RAID1 could approach RAID0 for reading data, but in my case,
with a few large files rather than a typical fileserver situation with lots of
concurrent access for many files (workstation vs server) I have a feeling RAID1
will not perform as well. 


We'll see how it goes...


Cheers,

   Brent