RAID: identical disks?

[J. Clarke]

Why would software RAID outperform hardware RAID?

Fast processor vs slow processor. "Hardware RAID" is generally implemented
in firmware on a Intel 960 or something else in that general performance
range, and those processors are not terribly fast.

 

[Curious George]

Fast processor vs slow processor.

It's a lot more complicated than that. For instance no mirror levels
require a fast processor and a poorly engineered parity level will
perform badly even with a fast processor.

"Hardware RAID" is generally implemented
in firmware on a Intel 960

an ancient IO processor. not really relevant today.

or something else in that general performance
range, and those processors are not terribly fast.

So by your logic in the HPC & enterprise NAS & SAN market the
best/fastest arrays use software on the host while the low-end entry
level models use hardware controllers.

There's more to it than that, like the processors today aren't that
slow, and true hardware raid off loads work from the main processor
(which can make things faster), unless this is a storage-only server
(which means it often doesn't matter much), and these IO processors
don't need to be very fast in most cases esp when there are no parity
calculations, and better controllers use NVRAM and sophisticated
optimizations which benefit performance, & like how different OS raids
perform differently, How some OS raids are more likely to trip an
unnecessary rebuild (which jeopardizes data & slows the system) etc
etc etc.

We could be here all day talking about how sometimes one is faster or
more reliable than the other. It depends on too many things across
too many raid levels and too many classes of products such an extreme,
error-riddled over-generalization can't possibly deal with.

 

[Folkert Rienstra]

What good would identity do if a disk dies after some time? Would
you have to replace every disk in the array? Pretty stupid idea IMO.

Whatever it is that you took 'identity' to be.

And no, RAID software is not tuned for particular disks. It takes
just the best each disk can give it and if it is smart it does
parallel requests to a limited degree. No disk dependence for
software optimisations in here.



I have: Two disks in the same Linux software-RAID 1

Should work fine but no Raid0 read speeds to be expected if on
same channel (except sequential reads). Writes go to cachebuffer,
so no access times to wait for in completing the transfers.

or 5

Very different kettle of fish.

on the same IDE channel. Terrible performance. The solution is however
not to change the disks, but to give each one its on IDE channel,

since IDE switchover in one channel is slow.

Clueless.
It's not the switchover, it's the combined access time that makes it slow
whereas on different channels only the longest access time sets the pace.

 

[Folkert Rienstra]

Why that? Do you assume array creation only with the primay disk
connected? Or is some RAID software too stupid to recognize which
disk is the smaller one?


This sure makes sense if both disks have a shared weakness, e.g.
extreme sensitivity to heat, power-spikes, shock or the like.
I find it overly paranoid for everyday use,

but especially in a machine you cannot easily get to this makes sense.

No it doesn't.

There you might even consider using 3 or more different disks in RAID1
array.

Which will all die eventually. That's relying on luck. That's paranoid.
The way to go is to avoid those 'weaknesses' to begin with and concentrate
on what you don't have a say in: premature death by normal use. Two drives
will suffice if you use some kind of an alarm system if the raid is broken.

Pretty unlikely without a common factor.

Pretty unlikely with a common factor too, with that 4 hours difference.

Maybe overvoltage or mechanical shock?

Maybe if several of those incidents occurred within those 4 hours.

 

[Folkert Rienstra]

Why would software RAID outperform hardware RAID?

It shouldn't but unfortunately reality shows different in some cases.
And it doesn't necessarily apply to all RAID types.

 

[Gerhard Fiedler]

It shouldn't but unfortunately reality shows different in some cases.
And it doesn't necessarily apply to all RAID types.

That makes sense... and is the same thing as it always has been. If you
need performance, it isn't enough to say "I go with RAM type X" or "I go
with architecture Y", you need to look at specific implementations,
devices, mobos, etc., how well it is implemented.

Gerhard