Re: [tlug] SATA software RAID or SAS hardware RAID?

[Curt Sampson <cjs@example.com>]

On Mon, 18 Dec 2006, Stephen J. Turnbull wrote:

> That may have been what you intended to do, but your words did
> something quite different.  In fact they argued from "Google and
> others" that there is no argument for specialized hardware at all.

Pretty much, especially at the low end. Let's summarize:

More drives can always be used to increase the reliability of a system,
assuming you keep up on the maintenance (i.e., replace drives when they
go bad). A pair of mirrored IDE drives and a hot spare will be more
reliable than a pair of mirrored SCSI drives that individually have
higher reliability. That configuration will generally be cheaper, too.

Custom controllers are already expensive, need to be spared, and can
still leave you with a lot of downtime. And worse yet, it's harder
finding the expertise to help you out when they go weird, since the
market for them is much smaller than that for open-source software RAID.
And they do go weird; I've seen it many times.

Custom controllers don't help with speed at all, except on extremely
heavily CPU-bound systems, and even there, it's usually cheaper to
upgrade the CPUs.

More drives add more disk arms, which is more important than anything
else (given the usual range of drives available today) in increasing
speed. Four 300 GB 7200 RPM IDE drives will be about as cheap as,
perhaps even cheaper than, two 300 GB 10,000 RPM SCSI drives, and will
be significantly faster.

Anyway....

cjs
--
Curt Sampson       <cjs@example.com>        +81 90 7737 2974