16 Jul
2020
16 Jul
'20
10:31 a.m.
On Thu, Jul 16, 2020 at 08:52:16AM -0700, Ali via cctalk wrote:
[...]
This is an article (for the layman) written in 2010
predicting the lack of
usability of RAID 6 by 2019:
www.zdnet.com/article/why-raid-6-stops-working-in-2019/. I found the math in
it interesting and the conclusions pretty true to my experience.
The author screwed up his maths and also made faulty assumptions.
The article states that "SATA drives are commonly specified with an
unrecoverable read error rate (URE) of 10^14. Which means that once every
200,000,000 sectors, the disk will not be able to read a sector." and then "2
hundred million sectors is about 12 terabytes." It seems he is using a sector
size of 64kiB. Standard SATA disks have 4kiB sectors.
"At that point the RAID reconstruction stops". Maybe on his garbage hardware
RAID controller with 64kiB stripes which chokes on a single-bit error in a
stripe because it's too dumb to figure out which disk is lying. ZFS is somewhat
smarter than that.
I am wondering if SW RAID is faster in rebuild times
by now (using the full
power of the multi-core processors) vs. a dedicated HW controller (even one
with dual cores).
Not only is software RAID faster now, but this has been the case for at least
15 years. The necessary calculations are trivially vectorisable and are usually
limited by memory bandwidth. Which is several orders of magnitude faster than a
hard disk.