Re: Soft raid 10

On Fri, Jan 26, 2007 at 04:05:25PM -0600, unit3 wrote:
> Dave Hall wrote:
> > Further evidence to dispel the myth that RAID is a performance solution.
> I'm not sure what you're talking about, since Steven's RAID benchmarks
> were much faster than the single disk ones, notably going from ~79MB/sec
> block write to ~148MB /sec, and (more importantly for server usage) ~255
> seeks/sec to ~650 seeks / sec.

Just adding an annecdotal data point for the hardware option Steven mentioned
he might consider if a software solution didn't perform good enough.

>
> Were we looking at the same benchmark results?
> > I did some annecdotal measurements a few years ago comparing a 7200RPM IDE
> > disc to a hardware RAID 5 using 10K RPM SCSI drives. The single IDE disc
> > consistently out performed the RAID. The same RAID hardware performed much
> > better running Linux (2.4 kernel) than Windows 2000 but not as good as the
> > single disc.
> That's because RAID5 computes hashes for each disk strip, which is very
> CPU intensive. RAID10 doesn't.

Computationally intensive but hardware RAID shouldn't be using any more CPU
than a plain old disc controller.
Received on Fri Jan 26 16:15:45 2007