Re: Re: Re: Re: RAID bus controller: Promise Technology, Inc. PDC20376 (FastTrak376)(rev02)
Len Sorensen wrote:
With the current trend of "win*" devices with more and more loading
firmware by driver, and off loading processing to the CPU all in the name
of cutting manufacturing costs (and wasting the end users expensive cpu)
I doubt we are going to get real raid chips on the mainboard ever.
people even seem willing to buy sound cards that don't off load any of
the signal processing while spending lots of money on the video card to
get higher frame rates in their games (while the sound card just ate 10%
of the cpu and framerate because it is cheap junk).
As long as the average computer buyer is clueless about computers (and
why shouldn't they be, it's a rather complicated piece of electronics),
there will continue to be companies making cheap junk that has impresive
looking specs for less money than something better.
Does anybody use any of the RAID features on any motherboard for Linux?
I prefer to use something I can trust, like md raid in linux. At least
then I get the source code to the drivers for my raid. If I have to
waste cpu cycles running a software raid, at least I want to know how it
works, and I also suspect the linux software raid is more cpu efficient
than whatever the proprietary software raid makers have put in their
I expect your suspicions about CPU efficiency for quality raid programs
like md are probably accurate. I would like to know with actual tests
except I can't run a comparison between raid software and raid chips
under Linux because my raid chips appear to be unsupported.
I agree with your preference for saving the CPU cycles with better
peripherals. The cheap integrated audio chip on my motherboard used 45%
of the CPU running mpg321 when my emu10k1 uses less than 1% with better
sound. Some motherboards now have the Creative chip on-board. My Radeon
produces 2100 fps using the GPU chip for OpenGL using only 15% of the
CPU while the software OpenGL produces only a 200-300 fps using almost
all of the CPU cycles. Even my Postscript printers have brains and use
none of the CPU at all to make a raster. I admit intelligent chips can
produce better results without wasting the CPU cycles I want instead.
So I wonder whether my raid chips are smart_raid or dumb_raid and
whether they can save the CPU cycles or not while remaining error free.
I expect the raid chips will actually improve in the future and wonder
when they will arrive. Some of the older SCSI RAID controllers were
probably fairly good and the newer SATA RAID controllers should have
been introduced with at least that capability.
I am afraid I do have cheap junk instead and so md raid is still the