Well, maybe on a small and some midrange system, but for large systems,
striping accross controllers is a very common way to do things.

Hardware RAID may be more convenient to set up but how many people out
there have heard of a hardware RAID system quietly eating the hot-spare
drive while the sysadmins were somewhere else?  I know of at least three
instances of that happening, the results of which were pretty bad.

Once again, I feel that the situation has been confused by whatever
windoze has for a software raid driver.  Linux has something different,
it's integrated deeply within the kernel, has access to all of the system
resources and is very fast.  It can use multiple controllers, so if you
lose the controller, you're not down.  This is how the commercial UNIXes
do things, and it's nice to see Linux getting there, too.

A decent RAID controller is fine for a small system, but it's far too
much of the system depending one one part for any large system.  The
minor inconveneince of when a disk has to be replaced and synchronizing
takes a tiny bit more system resources is vastly outweighed by the
increase in performance of using the kernel's built-in multiple device
subsystem.  The 'software' raid system is an enterprise-level feature,
that a home user can also take advantage of.

If anyone is interesetd, I can give a little talk at the next meeting
about high availability meets high performance.  I think that it would
be fun.

Let me know,

Tony Hammitt

Grrr...  Stupid fingers..  I sent this to kcug.org the first time.  I
wonder what their opinion is...

mike neuliep wrote:
> 
> Tony, one thing to remember when considering hardware vs software raid:
> 
> 1) Your statement is true that a P3 is faster and has more bandwidth than a PCI bus.
>       but...
> 
> 2) Raid controllers do everything onboard.  The PCI bus doesn't get hit up in a
> failmode situation.  The PCI bus only sees the normal read/write traffic
> 
>       and...
> 
> 3) In software raid, the PCI bus will even be more busy.  In addition to normal
> reads and write to the disks, it also has to handle the bandwidth for parity
> calculations in a failed mode situation.
> 
>         not relative to performance
> 
> 4) Hardware raid is more convinient.
> 
> Does anyone see any gaping flaws in my logic that with a decent RAID controller
> that it should always be faster in a failed situation or rebuild situation vs
> a software raid driver?
> 
>         Mike Neuliep

-- 
Doubt is not a pleasant condition, but certainty is absurd. -- Voltaire