Disk IO benchmarking

Ben Scott dragonhawk at gmail.com
Thu Jan 17 11:43:56 EST 2008


On Jan 17, 2008 11:14 AM, Mark Komarinski <mkomarinski at wayga.org> wrote:
> Depending on the CPU speed, you may find that software RAID is faster
> than hardware RAID as we did way back when.

  Yes.  If you have multiple CPU cores and one or more of them are
usually idle (common on many systems these days), RAID done in the OS
can be a real performance winner against special-purpose hardware,
especially compared to cheap RAID hardware.

  If you have a single-core system or a complex workload, though,
beware of being mislead by artificial benchmarks run on an unloaded
system.  If the system is doing nothing but running the benchmark,
there's lots of computrons available for RAID.  But if the system's
normal workload will be keeping the CPU(s) very busy, software RAID
may drag performance down.

> We stuck with a hardware card mostly for ease of setup, administration, and how
> easy it was to locate and replace failed drives ...

  I typically prefer hardware RAID for the same reasons.  Especially
booting under fault conditions.  I can never really bring myself to
trust that with software RAID, given all the weird problems I've seen
with boot loaders over the years.  (But then, I've always worked on
very heterogenous hardware.  If I had a stable platform to test
against, I might have higher confidence.  But when every BIOS has its
own unique failure modes...)

-- Ben


More information about the gnhlug-discuss mailing list