|
I. Introduction
RAID
is an acronym for “Redundant Array of
Independent (or Inexpensive) Disks”. It
refers to a set of methods and algorithms for
combining multiple disk drives as a group in
which the attributes of the multiple drives are
better than the individual disk drives. RAID can
be used to improve data integrity (risk
of losing data due to a defective or failing
disk drive), cost, or performance.
The different RAID implementations that are
available today offer different tradeoffs
between these three factors.
II. Origins of RAID
The concept of RAID was first defined in 1988,
when a group of computer scientists at the
University of
California Berkeley, (David Patterson, Garth
Gibson, and Randy Katz) published a paper
entitled “A
Case for Redundant Arrays of Inexpensive Disks
(RAID).”
The group observed that computer CPU speed and
memory size was growing exponentially, while I/O
performance was increasing at a much slower
rate. Unless I/O performance could be
significantly improved, computer systems would
not be able to take full advantage of the
rapidly increasing CPU and memory performance.
At the time, hard drive manufacturers addressed
this issue by designing and building Single
Large Expensive Disks (SLED). While storage
capacities of these disk drives were sufficient
for the times, I/O performance was still not
keeping up as the inherent mechanical
limitations of the hard drives were
significantly slower when compared to electronic
circuitry.
To overcome these limitations, the UC Berkley
scientists proposed that instead of storing all
data on one disk drive (with only one spindle),
why not combine several small inexpensive disks
(with many spindles) and stripe the data
(split the data across multiple drives), such
that reads or writes could be done in parallel.
To simplify the I/O management, a dedicated
controller would be used to facilitate the
striping and present these multiple drives to
the host computer as one large logical
drive. They estimated the performance
improvements would be an order of magnitude
greater than using SLEDs.
The problem with this approach was that the
small inexpensive PC disk drives of the time
were less reliable than the SLED’s. An artifact
of striping data over multiple drives is that if
one drive fails, all data on the other drives is
rendered unusable. It would be analogous to
deleting every
3rd
or
4th
sentence out of a book, then not knowing what
sequence the sentences were written in. To
compound this problem, by combining several
drives together, the probability of one drive
failing increases dramatically.
To overcome this pitfall, the scientists
proposed adding extra drives to the RAID group
to store redundant information. The thought was;
if one drive failed, another drive within the
group would contain the missing information,
which could then be used to regenerate the lost
information. Since all the information was still
available, the end user would never be impacted
with down time and the rebuild could be done in
the background. If users requested information
that had not already been rebuilt, the data
could be reconstructed on the fly and the end
user still would not know about it.
III. Original RAID Levels
The group outlined six RAID architectures
(levels) ranging from “Level 0 RAID” to “Level 5
RAID”. These levels provided alternative ways of
achieving storage fault tolerance, increased I/O
performance and true scalability.
They used three main building blocks in their
architectures:
1. Data Striping
-
Data from the host computer is broken up into
smaller chunks and distributed to multiple
drives within a RAID array. Each drives storage
space is partitioned into stripes. The stripes
are interleaved such that the logical storage
unit is made up of alternating stripes from each
drive. Major benefits are I/O performance and
the ability to create large logical volumes.
Used in RAID 0.
2. Mirroring.
Data from the host computer is duplicated on a
block-to-block basis across two disks. If one
disk drive fails, the data remains available on
the other disk. Used in RAID levels 1 and 1+ 0.
3. Parity.
Parity. Data from the host computer is written
to multiple drives. One or more drives are
assigned to store parity information. In the
event of a disk failure, parity information is
combined with the remaining data to regenerate
the missing information. Used in RAID levels 3,
4 and 5.
RAID 0 plus the five original RAID levels
developed by the Berkley scientists (along with
the RAID group’s performance, reliability and
cost assumptions) are listed as follows:
RAID Level 0
• Striped Disks.
o Performance: Very Good
o Reliability: Poor— Less than a single disk
drive (Non-Redundant)
o Cost:: Low
RAID Level 1
• Mirrored Disks.
o Performance: Slightly better than a single
drive
o Reliability: Excellent.
o Cost:: High (must purchase 2X disks).
RAID Level 2
• Uses bit interleaving and ECC. (This feature
is built in to most modern Disk drives now)
o Performance: Same as RAID level 1 for large
I/Os.
On small I/Os it is very bad; have to read all
disks; no parallelism.
o Reliability: Good
o Cost:: Cost is better than mirroring with 20%
to 40% cost overhead
RAID Level 3
• Uses byte interleaving with parity instead of
ECC. Parity data is stored on a dedicated drive.
o Performance: Same as RAID level 0 for reads,
Writes are slightly slower.
o Reliability: Good
o Cost:: Cost is 1 additional disk per RAID
group
RAID Level 4
• Same as RAID Level 3, but use sector
interleaving instead of bit interleaving.
o Performance: Same as RAID level 0 for reads,
Writes are slightly slower.
o Reliability: Good
o Cost:: Cost is 1 additional disk per RAID
group
RAID: Level 5
• Same as RAID Level 4, but distributes stripe
parity across all disks.
o Performance: Writes are slightly faster than
RAID 3 and 4, but reads tend to be
considerably slower.
o Reliability: Good
o Cost:: Cost is 1 additional disk per RAID
group
IV.
RAID
Today
The most widely used RAID levels today in 2002
are:
RAID 0:
Stripping
RAID 1: Mirroring
RAID 10 Mirroring and Striping
RAID (0+1) Striping and Mirroring
RAID 3: Striping with Dedicated Parity Disk
RAID 5: Stripping with Distributed Parity
As noted eailier, the different RAID levels
offer
a
variety of performance, data availabilfty, and
data integrity depending on the specific I/O
environment, however ft is important to
remember:
• RAID levels are not progressive
-
In other words, increasing the RAID level from 0
to 1 to 2 to 3 etc. does not give progressively
better data integrity, performance, or cost.
Each RAID level is independent and the numbering
is arbitrary (use of the term RAID
level
creates some confusion).
• Not all
RAID levels are redundant-
RAID 0 provides no data redundancy, in fact, it
is more prone to data loss than individual disk
drives, because if any drive fails in a RAID 0
group, all data is lost.
• There are no standards for RAID
-
Each vendor has its own implementations, and may
use different
terminology. Some vendors have invented their
own RAID terminology (e.g., EMC’s RAID-S and
Storage
Computer’s RAID 7). Vendors who claim to
implement RAID 3 are actually implementing a
modified RAID
3. Combinations such as 10, 0+1, and 53 are all
vendor defined. Storage users must be closely
examine
RAID implementations.
• RAID can be implemented in various places
in the computer system
-
The storage devices, the Host Bus Adapter (HBA)
and the host operating system (e.g., Windows
2000) can all implement RAID. It is possible to
use a combination of these, for example, RAID 0
(striping) in the storage array combined with
RAID 1 (mirroring) in the operating system. Each
location has benefits and shortcomings, which
need to be, understood by computer system
architects.
• Physical vs. Logical drive numbering-
Physical numbering refers to the physical
components in the storage array. Logical (or
virtual) numbering refers to the “disks” or
“volumes” that the host operating system “sees”
in the storage device. These two can be very
confusing to new storage users.
• Logical disks do not always map I-to-I with
physical disks
-
In RAID, several physical disk drives (or
portions of several physical drives) can be
grouped into a logical disk or Logical UNit
(LUIV).
Each LUN can be broken into logical blocks of
512 bytes each, numbered 0 through “n” (the
Logical Block Number or
LBIV).
For example, a 100GB LUN has approx. 200,000,000
logical blocks.
• Logical volumes are
very similar to logical drives
-
A logical volume is composed of one or
several logical drives, the member logical
drives can be the same RAID level or different
RAID levels.
The logical volume can be divided into
partitions. During operation, the host sees a
non-partitioned logical volume or a partition of
a partitioned logical volume as one single
physical drive.
V. RAID Summary
RAID can be a powerful tool in a storage
environment. Using a RAID storage subsystem has
the following advantages:
• Provides fault-tolerance by mirroring or
parity operation.
• Increases disk access speed by breaking data
into several blocks when reading/writing to
several drives in parallel.
• Simplifies management by weaving multiple
drives together to form a large volume or groups
of volumes.
Today, the major RAID levels available offer the
following characteristics:
A. RAID
0: Striped Disk Array
Raid 0 is not a fault tolerant RAID solution, if
one drive fails, all data within the entire
array is lost. It is used where raw speed is the
only (or major) objective. It provides the
highest storage efficiency of all array types.
Pro’s
• Improved I/O performance
• Most capacity-efficiency RAID level
• Ability to create large logical volumes.
Con’s
• RAID 0 does not utilize disk space for
redundancy.
• If one disk fails, all data within the stripe
set is lost.
Configuration
• RAID 0 array’s are made by grouping two or
more physical disks together to create a virtual
disk and making this virtual disk appear as one
physical disk to the host. Each physical drives
storage space is partitioned into stripes. The
stripes are then interleaved so that the virtual
disk is made up of alternating strips
from each drive.
• To increase performance, RAID 0 writes block
level data across all available stripes in the
RAID 0 group, enabling parallel disk I/O, which
optimizes I/O performance.
• Ideally, the size of the stripe is large
enough to fit one record. The record broken into
smaller sizes and evenly distributed across all
drives in the stripe group.
Uses of RAID 0: RAID 0 should be used
with applications that require the highest level
of performance and use non-critical or temporary
data, such as:
• Full motion video editing applications
• Prepress editing applications
• Scratch files for CAD
• Any application where the original content is
backed up and can be easily restored. The time
saved in doing normal data processing work with
RAID 0 more than makes up for the time lost in
infrequent disk crash events.
B. RAID 1: Mirrored Disk Array
RAID 1 provides complete protection and is used
in applications containing mission critical
data. It uses paired disks, where one physical
disk is partnered with a second physical disk.
Each physical disk contains the same exact data
to form a single virtual drive.
Complete data protection is achieved by
simultaneously writing two exact, block level
copies, of data to each disk in a mirrored pair.
There is no striping. Read performance is
improved since either disk can be read at the
same time. Write performance is the same as for
single disk storage. RAID-i provides the best
performance and the best fault-tolerance in a
multi-user system.
Wfth RAID 1, the host will see what it believes
to be a single physical disk of a specific size.
(The host does not know or care about the
mirrored pair) The RAID controller manages where
the data is written and read. This allows one
disk to fail without the host ever knowing,
providing time for service personnel to replace
the failed drive and initiate a rebuild.
Pro’s:
• Highest level of protection
-
Mirroring provides 100% duplication of data.
• Read performance is faster than a single disk;
(if the array controller is capable of
performing simultaneous reads from both devices
of a mirrored pair)
• Delivers the best performance of any redundant
array type during a rebuild.
o No re-construction of data is needed. If a
disk fails, copying on a block by block basis to
a new disk is all that is required.
o No performance hit when a disk fails; storage
appears to function normally to outside world.
• The only choice for fault tolerance, if only
two drives are used.
Con’s
• Raid 1 writes the information twice, because
of this there is a minor performance penalty
when compared to writing to a single disk.
• I/O performance in a mixed read-write
environment is essentially no better than the
performance of a single disk storage system.
• Requires two disks for 100% redundancy;
doubling the cost.
Uses of
RAID
1:
RAID 1 provides the most complete protection,
however, it also requires duplication of
physical disks. In the past, this RAID level was
used exclusively in smaller mission critical
networks to keep costs down. As the cost of
storage arrays decline, many system architects
are reconsidering the use of RAID 1 in larger
applications. Typically, these applications
involve mostly read-only operations or light
read-write operations. An example of a typical
RAID 1 implementation is a data entry network.
It is recommended for applications where:
• Data availability is very important
• Speed of read access is very important
• Read activity is heavy
• Applications needing logging or record keeping
C. RAID 10: Mirroring and Striping
RAID 10 consists of multiple sets of mirrored
drives. These mirrored drives are then striped
together to create the final virtual drive. The
result is an extremely scalable mirror array,
capable of performing reads and writes
significantly faster (since the disk operations
are spread over more drive heads).
Pro’s:
• Very high reliability
o Because there are multiple mirror sets, this
configuration can actually handle multiple disk
failures and still survive (*with one
exception).
• Provides highest performance with data
protection
• By striping multiple mirror sets, RAID 10 can
create larger virtual drives. The host computer
will see what it believes to be a single
physical disk of a specific size.
• Can be tuned for either a request-rate
intensive or transfer-rate intensive environment
*Disk failures occurring within the same mirror
set are the exception which is extremely rare.
Con’s:
• Like Raid 1, RAID 10 writes the information
twice, because of this there is a minor
performance penalty when compared to writing to
a single disk.
• I/O performance in a mixed read-write
environment is essentially no better than the
performance of a single disk storage system.
• Requires an additional disk to make up each
mirror set.
Uses of RAID 10:
Applications where high performance and
reliability are paramount are ideal for RAID 10.
Examples would be on-line transaction
processing environment and financial
transaction processing environment. It is
recommended for applications where:
• Data availability is critically important
• Overall performance is very important
0. RAID
(0+1): Striping and Mirroring
Not to be confused with RAID 10 (they are very
different). Raid 0+1 flips the order of RAID 10.
Drives are first striped, then these drives are
mirrored. Typically, two or more disks are
striped to create one segment and an equal
number of drives are striped to form an
additional segment. These two striped segments
are then mirrored to create the final virtual
drive.
Pro’s
• High I/O performance
• Ability to create large logical volumes
Con’s
• Reliability is less than RAID 1 and 10. If one
disk fails you essentially, have a RAID 0
configuration. Due to the multiple disks that
make up the RAID 0 segment, the probability of a
disk failure is greater.
• Requires duplicate drives. Capacity of
physical drives is half.
Uses of RAID 0+1:
Applications that require high performance,
but are not overly concerned with achieving
maximum reliability.
E. RAID 3: Striping with Dedicated Parity Disk
RAID 3 is a fault tolerant version of RAID 1
(Striping). Fault tolerance is achieved by
adding an extra disk to the array and dedicating
it to storing parity information. Parity
information is generated and written during
write operations and checked on reads. It
requires a minimum of three drives and provides
data protection.
In the event of a disk failure, data recovery is
accomplished by calculating the exclusive OR (XOR)
of the information recorded on the other drives.
Since an I/O operation addresses all drives at
the same time, RAID-3 cannot overlap I/O.
Forthis reason, RAID-3 is best for single-usersystems
with long record applications
Pro’s
• Good data protection
• Good write performance
• Good read performance
• The amount of useable space is the number of
physical drives in the array minus 1.
Con’s
• A single disk failure reduces the array to
RAID 0
• Performance is impacted when degraded
• Poor performance with small data transfers.
• Limited to single user environments.
Uses of RAID 3:
This version of RAID is best suited for:
• Single user, single tasking environments with
large data transfers.
• Heavy write applications.
• Large volumes of data are stored
F. R1JD
5: Stnping
and Panty
Raid 5 is similar to RAID 3 but the parity is
not stored on one dedicated drive, instead
parity information is interspersed across the
drive array. RAID 5 requires a minimum of 3
drives. One drive can fail without affecting the
availability of data. In the event of a failure,
the controller regenerates the lost data of the
failed drive from the other surviving drives.
By distributing parity across the arrays member
disks, RAID Level 5 reduces (but does not
eliminate) the write bottleneck. The result is
asymmetrical performance, with reads
substantially outperforming writes. To reduce or
eliminate this intrinsic asymmetry, RAID level 5
is often augmented with techniques such as
caching and parallel multiprocessors.
Pro’s:
• Best suited for heavy read applications.
• The amount of useable space is the number of
physical drives in the virtual drive minus 1.
Con’s
• A single disk failure reduces the array to
RAID 0
• Performance is slower than RAID 1 when
rebuilding
• Write performance is slower than read (write
penalty)
• Block transfer rate is equal to single disk
rate
Uses of RAID 5:
RAID 5 is a general-purpose RAID storage
solution. It is recommended for applications
where:
• Data availability is important
• Large volumes of data are stored
• Multi tasking applications using I/O transfers
of different sizes
• Good read and moderate write performance is
important
G. Comparing RAID configurations
Each of the described RAID levels offers
different characteristics in terms of cost and
performance. The following table compares the
cost, data availability, and
I/O
performance of the commonly known RAID levels.
I/O performance is shown both in terms of large
I/O requests, or relative ability to move data,
and random I/O request rate, or relative ability
to satisfy I/O requests. Since each RAID level
has inherently different performance
characteristics relative to these two metrics
1
The data transfer capacity and I/ 0 request rate
columns reflect only I/O performance inherent to
the RAID model, and do not include the effect of
other features, such as caching.
H. RAID Set-up Considerations
Setting up a fault tolerant RAID array involves
trading off economy for MTDL (Mean time to Data
Loss). MTDL is the probable time to failure for
any component that makes data inaccessible.
If your data is backed up and performance and
cost are your primary concern, then RAID 0 is
the logical choice.
If you determine that you need data protection,
then you have two choices to protect your data,
Parity or Mirrored arrays.
The “cost” of data protection for a Parity RAID
array is the equivalent of one disk per RAID
group. At first glance, it would seem to be a
no-brainer to select Parity Raid over Mirrored
RAID
• When a drive fails in a parity RAID array
(RAID 3 or 5), the array becomes a RAID 0-stripe
group. If a second drive failure occurs before
the array completes a rebuild, then you loose
all data within the array.
If you are comfortable with this, then Parity
RAID is probably the most economical solution
for you. It would also seem to be very cost
effective to build parity arrays with several
disks, however, consider the following when
configuring your parity array:
• More disks in a parity RAID array affects
write performance adversely.
• More disks in any RAID array increases the
probability of a drive failure,
• Modern disk drives can be as large as 160 GB.
By creating a parity array with several disks,
the capacity of the array skyrockets,
dramatically increasing resynchronization time
after a disk failure. This has a major impact on
array performance and forces, the array to run
“unprotected” for an extended period.
If you determine that the performance and/or
protection limitations of parity RAID are too
great, then a Mirrored (RAID 1) array or a dual
level array (such as RAID 10) should be your
choice.
VI. Close
Selecting the proper RAID level and setup for a
disk storage array is the key to properly
balancing system costs and performance needs.
This primer and its overview are an excellent
point to properly starting your RAID storage
selection activities. Should you have any
further questions the professionals at 9 to 5
Computer are ready to assist you in selecting a
RAID system and setup to best meet your needs
and goals.
|
