WriteLog
[Top] [All Lists]

[WriteLog] Of RAID, Drives, warranties, and reliability... (long)

To: <writelog@contesting.com>
Subject: [WriteLog] Of RAID, Drives, warranties, and reliability... (long)
From: aaron.hsu@unistudios.com (Hsu, Aaron)
Date: Wed Jan 29 06:12:02 2003
My 2 cents...

RAID - Redundant Array of Inexpensive Disks.  Often (incorrectly), the word 
"Inexpensive" is substituted with "Independent".  I believe this error first 
started when a well known computer magazine published an article with the word 
"Independent" and the confusion began.  As stated before, other devices can 
also be configured "RAID-like".  One common example are RAID Tape arrays.  I've 
seen multi-drive terrabyte LTO arrays at conventions.  Impressive!

RAID levels 0, 1, and 5 are most common.

RAID 0 - Stripping across two or more drives with no redundancy (isn't this an 
oxy-moron? <g>).  RAID-0 allows you to combine several drives of the same 
capacity and present them to the system as one logical drive.  Somewhat risky, 
but fast.  When properly configured with drives of known quality (e.g. 
Maxtor/Quantum Atlas 10K III series) and with a backup system in place, RAID-0 
is hard to beat.

RAID 0 - JBOD.  Typically, all RAID configurations use drives of the same 
capacity.  If you use different capacity drives, then the RAID controller 
treats all drives as the same capacity of the smallest drive in the array.  For 
example, if you have two 10GB drives and a 4GB drive, then the entire array is 
treated as three 4GB drives (12GB RAID-0).  Some controllers support JBOD - 
"Just a Bunch Of Disks".  This RAID-0 implementation allows the full capacity 
of the drives to be utilized - so in the above example, the array would have a 
total capacity of 24GB.

RAID 1 - Drive mirroring.  Data is written identically to two volumes.  If one 
volume fails, data can be retrieved from the other volume.  In higher-end 
systems, the failed drive can be "hot-swapped" (replaced) without powering down 
the system while the other drive services requests.  Replacement drive data is 
re-built in the background.

RAID 1 - Duplexing.  Same concept as mirroring, but drives are on two separate 
controllers.  Duplexing basically adds a redundant controller.

RAID 5 - Data striping (distributed parity).  The data is written in "stripes" 
(or blocks/chunks/etc) across 3 or more drives.  The "parity data" is also 
striped across all the drives.  This speeds up READ requests as drive heads can 
be positioned to a "ready" state other drives while another drive is actually 
transferring data (lowers total "Latency" times).  May also speed up writes, 
but often negated by read-after-write verify integrity checks.  The entire 
array can tolerate a failure of one or two drives, depending on the total 
number of drives in the array, but the array will operate much slower until the 
failed drive is replaced and the data rebuilt.  Failed drives can be 
"hot-swapped" in supported systems.  The total capacity of the array is 
"Drivesize * (N-1)" where "N" is the number of drives in the array.  So, four 
10GB drives would have a RAID-5 capacity of 30GB.  RAID-4 is similar to RAID-5, 
but used a dedicated parity drive (RAID-5 is better).

One can also combine RAID configurations in the same system for additional 
redundancy.  RAID 0/1 (mirrored RAID-0 stripe array) and 1/0 (striped RAID-1 
mirror array) are popular (btw, 0/1 and 1/0 may look the same, but they're 
not...1/0 is better).  RAID 5/1 (mirrored RAID-5 array) is commonly found in 
mission critical enterprise-sized servers and storage arrays.  RAID 5/0 is also 
used where speed and capacity are important, but it's rare.

ZIP drive reliability and MTBF...good, convenient drive.  However, early ZIP 
drives suffered from a "click of death".  The drive would litterally commit 
suicide and sever it's R/W head trying to read a disk.  Iomega eventually fixed 
the problem, but the stigma remained.  The "click of death" is probably why you 
see so many refurb ZIP drives available for sale.  Also, be aware that the ZIP 
250 drive will correctly read a 100MB disk, but if you write to a 100MB disk, a 
100MB drive may no longer be able to read the disk.  This is due to R/W head 
size.  Similar problem existed with the old 5.25" 1.2MB floppy drive writing to 
360K disks.  I believe the new ZIP 750 has the same problem with 250MB disks 
(it will only read 100MB disks, not write).  As to a statement earlier about 
the ZIP drive's MTBF, I would say that it's actually *shorter* due to the 
repeatitive spin-up spin-down cycles - more strain on the motor.

Drive companies...All drive companies have had their lemons.  All companies 
have had their shining stars.  IBM recently had a lot of problems with their 
popular 75GXP(?) line.  My co-worker had one that failed after 3 months, got a 
replacement under warranty and it failed 3 weeks later.  A 2nd replacement 
drive also failed.  There are many more stories like this about this particular 
line of drives.  Western Digital did a recall a couple of years ago due to a 
faulty component.  Currently, Fujitsu is recalling some drives due to a faulty 
component (get this, some chip manufacturer used phospherous in the plastic 
used to make an IC chip and now they're failing.  Phospherous?  Isn't that 
corrosive???).  And Fujitsu had a track record of making some of the most 
reliable drives on the market.

Warranties - For a while, drives were near perfection and reliability was high 
so companies offered longer warranties as a marketing incentive.  Alas, as the 
HD market tumbled along with prices, the QC fell too.  Companies started 
replacing more drives under warranty than originally anticipated.  Combine that 
with the currenty economic conditions and drive companies are now shortening 
the warranties back to 1 year.  Beware...some of these warranties are from date 
of manufacture, not date of sale.  Also, in general, a drive company might have 
several different "lines" of drives.  The higher-end lines often come with a 
longer warranty.  Most "enterprise" class SCSI drives still come with a 5 year 
warranty (but cost quite a bit more!).

But, alas, I've digressed long enough.  Just though I'd offer some 
clarification to some topics recently brought up.

73,

  - Aaron Hsu, NN6O
    (athsu)@unistudios.com
    (nn6o)@arrl.net
    No-QRO Int'l #1,000,006



All contents of this message copywright Aaron Hsu, NN6O

<Prev in Thread] Current Thread [Next in Thread>