> My hw raid results are just one controller. You aren't measuring
> throughput when the controller is busy doing background raid initialization,
> are you? That would hurt performance.
Nope... before running bonnie, I make sure the drives are *totally* idle.
That's another benefit to only letting the h/w do raid0, as it doesn't
really have anything to initialize like it would with 1 or 5 (or at
least it seems that way, as the drives are idle).
> Unfortunately, at least with the DAC960 driver 2.2.1, there does not
> seem to be any indication of background initialization in /proc/rd/c0/*,
> so I have to look at the drive activity lights (and the performance)
> to tell when initialization is complete.
Same here.
> It isn't reporting 99% of 4 Xeons, but 99% of one Xeon.
> Bonnie just takes cpu/elapsed*100.
Hmmm I guess I should really be using /usr/bin/time to fetch
the system time as an interpretation of kernel overhead for
raid5? Probably too inaccurate or useless, so I guess I'll
get /proc/profile going and see what readprofile gets me..
James
--
Miscellaneous Engineer --- IBM Netfinity Performance Development
