|
The IBM RAID Controller provides the Server 95A with a
data redundancy technique that distributes data across all the drives in
the array. In the event of a single drive failure, read and write
requests are satisfied by the remaining disks in the array, without loss
of data. The defunct hard disk drive appears to be operating correctly,
if at reduced performance, despite its total failure.
When you replace the defunct drive, the data in logical
drives assigned levels 1 and 5 is rebuilt on the new hard disk drive using
the parity information stored on the other hard disk drives in the array.
It is important that the defunct drive be replaced and rebuilt in a timely
manner so that you avoid multiple-drive failure and reduced performance.
Once the drive contents are rebuilt, full performance and fault-tolerance
functions are restored.
The Server 95A supports RAID levels 0, 1, and 5. Following are
descriptions of these three RAID levels.
RAID Level 0
RAID Level 1
RAID Level 5
Rebuild Process For Non-Techo Dweebs
RAID Level 0
RAID level 0 stripes the data across all of the drives of the array.
Warning! Drive failure results in
loss of data in logical drives assigned RAID level 0.
This RAID level offers substantial speed enhancement, but
allows for no data redundancy; therefore, a hard disk failure within the
array results in loss of data in the logical drive assigned RAID level
0, but only in that logical drive.
Note: Although in this case
logical drives assigned RAID level 1 or 5 in the same array do not suffer
loss of data, the status of these drives is Critical and they cannot sustain
another hard disk drive failure. They operate at reduced performance,
and the defunct hard disk drive should be replaced and rebuilt promptly
to avoid a multiple hard disk drive failure.
When the defunct drive is replaced, all the logical drives assigned
RAID levels 5 and 1 are rebuilt onto that hard disk drive; the level-0
logical drive is defined, but the data in the level-0 logical drive is
lost.
Even though the risk of data loss is present, you might
want to consider assigning RAID level 0 to one of the logical drives so
that you can take advantage of the speed offered with this level.
One way you might use this logical drive is to enter data
that you back up each day and for which safety is not of primary importance;
that is, you can re-create the data easily.
Another instance when you might want to use a level 0
logical drive is when the work you are doing requires large capacity-level
0 provides you the largest capacity of the three RAID levels offered because
no room is taken up for data parity storage.
Notes:
1. You cannot define a hot-spare drive for an array that contains
a logical drive assigned RAID level 0.
2. All logical drives in an array containing only one drive are
automatically assigned RAID level 0.
Ed. To run individual
drives (one or two) on a RAID controller, define them as RAID 0 Array.
You are limited to two arrays on either the Passplay or Cheetah RAID controller.
RAID Level 1
RAID level 1 supported by the IBM RAID Controller provides
an enhanced feature for disk mirroring that stripes data and copies of
the data across all the drives of the array. The first stripe is
the data stripe, and the second stripe is the mirror (copy) of the first
data stripe but shifted one drive. Because the data is mirrored, the capacity
of the logical drive when assigned level 1 is 50 percent of the physical
capacity of the grouping of hard disk drives in the array.
If you have three or more drives in the array and you
want data redundancy, it's a good idea to choose level 5 first. To
learn why this might be to your advantage, see RAID Level 5.
RAID Level 1 Illustration:
This table shows data arranged in a disk array with three
hard disk drives, with the logical drive assigned RAID level 1. Notice
that the mirror of the first data stripe (in this case, XXX YYY ZZZ) is
shifted one drive. The other data stripes in RAID level 1, as shown
in this table, follow the same pattern.
DRIVE 1 DRIVE 2 DRIVE 3
Stripe XXX YYY ZZZ
Mirrored stripe ZZZ XXX YYY
Stripe AAA BBB CCC
Mirrored stripe CCC AAA BBB
The above illustrates three drives in an array, with one logical drive
assigned RAID level 1.
If you have only two drives available (for example, if
you define one of the drives that came with your server as a hot spare),
the second drive is an exact mirror copy of the first drive.
RAID Level 5
RAID level 5 stripes data and parity of the data across
all the drives of the array. The capacity of the logical drive when
assigned level 5 is reduced by one drive (for data parity storage).
Level 5 is generally the most desirable choice because
it offers both data protection and increased throughput. It gives you higher
capacity than level 1, but level 1 offers higher performance. If
after using level 5 you are dissatisfied with the performance and can tolerate
lower capacity, you can either redefine the level 5 to a level 1 (refer
to Redefining Space in an Array), or you can use a logical drive that you
have assigned level 1.
If you want to have a hot-spare drive and also assign
level 5, you must have at least four hard disk drives and no logical drives
assigned RAID level 0. To maintain capacity, the size of the additional
drive can be larger but must be no smaller than the size of the three drives
that came with your server. All the drives in an array are configured
to the capacity of the smallest.
RAID 5 Rebuild Process Explained for Non-Techno-Dweebs
Helmut P. Einfalt speaks in a way we can understand:
Actually, for six HDs (one hot spare) the data are chopped up into
chunks of user-definable size (4k, 8k, 16k) and written to the disks in
the following order:
HD1 HD2 HD3 HD4 HD5 SSS
111 222 333 444 CCC ---
CCC 111 222 333 444 ---
444 CCC 111 222 333 ---
333 444 CCC 111 222 ---
222 333 444 CCC 111 ---
111 222 333 444 CCC ---
CCC is the checksum, SSS in this case is the hot spare. That one isn't
accessed at all, and sits there idle. The sequence left-to-right or
right-to-left can be changed in the setup.
If anything goes wrong, say, with HD2 the situation will be:
HD1 HD2 HD3 HD4 HD5 SSS
111 ??? 333 444 CCC ---
CCC ??? 222 333 444 ---
444 ??? 111 222 333 ---
333 ??? CCC 111 222 ---
222 ??? 444 CCC 111 ---
111 ??? 333 444 CCC ---
At that point the RAID will automatically calculate the missing data
from "checksum minus available data" and insert the calculated data into
the hot spare:
HD1 HD2 HD3 HD4 HD5 HDS
111 ??? 333 444 CCC 222
CCC ??? 222 333 444 111
444 ??? 111 222 333 CCC
333 ??? CCC 111 222 444
222 ??? 444 CCC 111 333
111 ??? 333 444 CCC 222
The dead HD2 will be marked as dead, and the array will work on with:
HD1 DDD HD3 HD4 HD5 HD2
111 --- 333 444 CCC 222
CCC --- 222 333 444 111
444 --- 111 222 333 CCC
333 --- CCC 111 222 444
222 --- 444 CCC 111 333
111 --- 333 444 CCC 222
Spare now has become HD2, while the original HD2 is marked as defunct
(DDD). At that point, one should swap out the dead HD for a new one
that will in turn become hot spare. However, even if this is not done,
and one more drive crashes, the RAID will *still* be able to provide the
missing data on-the-fly:
HD1 DDD HD3 HD4 HD5 HD2 RAID
111 --- 333 ??? CCC 222 =444
CCC --- 222 ??? 444 111 =333
444 --- 111 ??? 333 CCC =222
333 --- CCC ??? 222 444 =111
222 --- 444 ??? 111 333 =xxx
111 --- 333 ??? CCC 222 =444
RAID in this case is the controller supplying the missing data. In the
penultimate case it is not required to do so, since only the checksum is
missing and that one isn't what the system wants anyway.
If something goes wrong at *that* instance, however, you're stuck...
I've run the last situation on my -B0C with 5x2,1 DFHS for a week before
I got a new HD -- no problem. And since the controller (Cheetah) is
pretty well equipped with its 4 MB cache there was no noticeable slack
in response speed.
The nice thing is that you're not limited to 6 drives -- you can go
up to 15 max on the Cheetah, which makes the "loss" proportionally smaller.
If I could get an SCSI enclosure to match the 9595-B0C (was it 3512?),
I'd set up my system that way and keep *all* data in there, not only the
critical ones...
|