On Sat, 15 Nov 2008, Richard Scobie wrote:

Mikael Abrahamsson wrote:

quoted

On Fri, 14 Nov 2008, Billy Crook wrote:

quoted

Obviously RAID-class drives benefit in hardware raid (with hardware
raid controllers).  But do RAID-class drives benefit when used on a
non-raid hardware controller, with linux software raid (mdadm)?  What
about a hardware raid controller in JBOD mode?

This is not a HW vs SW issue, it has to to with the behaviour of the drives 
in conjunction with the OS when a certain thing happens.

In a non-raid environment, you want to read the data at any cost, so 
waiting 2 seconds for it to read is no real problem, you just want the 
data.

In a raid environment you can reproduce the block by way of mirror or 
parity, so you want the read to fail after a short time so it won't delay
the reading of the data.

Here is a paper about the Western  Digital feature "TLER", which is used in 
all their RE series RAID drives.

http://www.wdc.com/en/library/sata/2579-001098.pdf

In addition, these drives are better able to cope with the problem of 
mechanical vibration/resonance that occurs when multiple drives are mounted 
in close proximity:

http://www.wdc.com/en/library/sata/2579-001079.pdf

These features will benefit HW and SW RAID implementations.

I have no association with Western Digital, other than as a satisfied user.

I have enabled TLER on all of my Velociraptors, we will see if they 'hang'
up anymore and drop out of the raid or just throw bad sector errors right
away, however, I am not sure md(Linux) can handle this, where as a 3ware card
remaps the bad sector, the kernel does not (or at least I have never seen it
do it on my system), instead the drive shows an Uncorrectable_Sector or 
similar..

Justin.