IBM serveraid and linux

Lennart Sorensen lsorense-1wCw9BSqJbv44Nm34jS7GywD8/FfD2ys at public.gmane.org
Thu Jan 6 15:06:43 UTC 2005


On Thu, Jan 06, 2005 at 01:09:13AM -0500, Henry Spencer wrote:
> On Thu, 6 Jan 2005, Dave Stubbs wrote:
> > ...Wouldn't it be faster?  Unless (I guess) the modern CPUs 
> > are just so powerful they can run the calc faster than the dedicated 
> > copro on the RAID card.
> 
> Remember that modern CPUs have a cycle time of a fraction of a nanosecond.
> That dedicated coprocessor probably doesn't have anywhere near that clock
> speed, as witness the fact that it doesn't have its own howling cooling
> fan.  Unless it's a custom design, which it probably isn't, it can't do
> that calculation anywhere *NEAR* as fast as the main CPU can. 

As far as I can tell, 3ware cards use a custom design for their raid
controller, and it seems to have very good performance (at least in
benchmarks), while the serveraid cards I have seen all had intel i960
(as far as I recall) chips on them, which is certainly not custom designed
for the purpose of raid.  After all many postscript printers use the
same chip, and I don't see a whole lot of similarity between postscript
rendering and raid.

> The data takes longer to get through the *wires* than the main CPU needs
> to run parity on it.  (An electric signal travels only a few centimeters
> along a wire in one main-CPU clock cycle!)
> 
> This has always been the problem with such I/O-card coprocessors:  unless
> they have a radically different architecture than the main CPU, so much
> better suited to the problem that they inherently handle it much more
> efficiently, they will always be much slower than the main CPU, so they're
> usually pretty useless.  By their nature, they can't command anywhere near
> the level of resources lavished on making the main CPU fast.  And the
> mind-boggling acceleration of CPU speeds in the last decade has made this
> problem even worse. 

Yeah raid cards need a dedicated xor engine, not a generic cpu.

Lennart Sorensen
--
The Toronto Linux Users Group.      Meetings: http://tlug.ss.org
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://tlug.ss.org/subscribe.shtml





More information about the Legacy mailing list