Thor Lancelot Simon writes:
> On Sun, Jan 04, 2004 at 06:46:35PM -0600, Greg Oster wrote:
> >
> > You mean like:
> > 
> > #define RF_PROTECTED_SECTORS 64L    /* # of sectors at start of disk to
> >                                        exclude from RAID address space */
> > 
> > in /usr/include/dev/raidframe/raidframevar.h ?  :)  (Sorry.. it's not 
> > been as advertised as well as it might have been...)
> 
> So, if I use RAIDframe over volumes that themselves have a "stripe size",
> such as hardware RAID units, all my values in the disklabel need to be
> shifted back down by 64 in order to fall on nice even stripe boundaries?

Sure, you might need to shift them one way or another for maximal 
performance.  I've had to do this on occasion for underlying RAIDframe 
devices too... 

> Given that 64K is a pretty common stripe size, but 64 _sectors_ is 32K,
> doesn't this basically give worst-case misalignment for many underlying
> devices?

At the time I picked 64 sectors, the most common "underlying device" was the 
simple disk, and I was concerned with a) missing any disklabels, and 
b) missing any other boot-related stuff (e.g. as found on hp300).  

My guess is that if you can afford the "big underlying hardware RAID 
devices" that you can also afford to play a bit with the offsets and 
maybe give up a *little* space to get the best performance :)

Later...

Greg Oster