EC/MIS 619

1
EC/MIS 619

• RAID Data Protection

2
RAID

• RAID a redundant array of inexpensive disks
• using multiple fixed-disk drives, high speed
  controllers and special software drivers to
  control the safety of your data and improve the
  performance of fixed-disk system.
• All commercial systems use Small Computer Systems
  Interface (SCSI pronounced scuzzy)

3

• RAID levels of 1 or higher protect your data by
  spreading it on multiple disks then calculating
  and storing parity bit information.
• This redundancy allows one drive to fail without
  causing the array itself to fail.
• RAID 1 increases disk subsystem performance by
  distributing data across several drives, allowing
  the same data to be retrieved from many locations
  - depending on which is closer to the read
  head(s).

4
RAID Levels

• Several level of RAID exists. RAID 0 increases
  read and write performance but does not provide
  data protection.
• RAID was initially designed for mainframe and
  microcomputers. Until recently its deployment was
  limited by price.
• Plummeting disk price has made RAID more common
  of late.
• Disk drives provide two functions read and
  write. RAID may be chosen to optimize these
  functions.

5
RAID 0

• High performance, zero redundancy array.
• Isnt truly RAID at all.
• Data is lost if one drive fails
• It strips blocks of data across multiple disks to
  improve subsystem throughput.
• RAID 0 is used for applications needing the
  highest possible read and write rates.

6
RAID 0 with 2 disks
Sector 1
Sector 2
Sector 3
Sector 4
Sector 6
Sector 5
Sector 7
Sector 8
Sector 9
Sector 10
Disk 2
Disk 1
7

• RAID 0 is important because the same stripping
  mechanism used in RAID is used to improve
  performance in other RAID levels.
• RAID 0 is inexpensive because
• no additional disk space is needed for parity
  data
• it uses simple algorithms that dont add much
  overhead or require a dedicated processor.
• RAID 0 uses stripping to store data
• data blocks are alternately written to different
  physical drives that make up the logical drive
  used by the array.

8

• Each read request is directed to the individual
  drive on which multiple blocks
• multiple reads request are generated serially
• stripping allows data transfer to occur in
  parallel.
• Overall read time is significantly reduced.
• Efficiency is influenced by size data blocks

9
RAID 1 with 2 disks
File Data
Sector 1
Sector 1
Sector 2
Sector 2
Sector 3
Sector 3
Sector 4
Sector 4
Sector 5
Sector 5
Disk 2
Disk 1
10
RAID 1

• We avoid loosing data by making copies of it.
• RAID 1 provides 100 redundancy- if a disk is
  lost in array1, theres another drive with an
  exact duplicate of the failed drive's contents.
• RAID 1 offers the highest level of redundancy
  (but at the highest cost).

11

• Mirroring- each drive has a twin drive. Write
  functions take place simultaneously.
• Disadvantages
• twice as many disk needed for storage
• slow writes - because of overheads introduces
  with the need to write to twin drives and
  maintain coherency of their contents.
• Advantages
• duplicates means data loss is less likely
• faster reads - reads can be made from drive whose
  head is closest to the data.

12

• Duplexing similar to mirroring but adds a second
  host adapter to control the second set of drives.
  
• Introduces cost of second adapter
• Duplexing with two cards eliminates the host
  adapter as a single point of failure.
• In a large server the cost of duplicating every
  disk gets very expensive. Cost are reasonable for
  smaller systems

13

• Read Performance is faster than that of stand
  alone drive.
• Offers two read alternatives
• Circular queue or round-robin scheduling reads
  are alternated between two physical drive with
  each drive handling every second request.
• Geometric, regional, or assigned circular
  scheduling overcomes slow reads by giving two
  drives the responsibility to cover only half of
  the physical drives, thus head positioning time
  is minimized.

14

• RAID 1 Write Performance is more problematic.
  Data has to be written twice (like going through
  the checkout line twice)
• This double write that facilitates the high level
  of data safety.
• Overall RAID 1 performance for most server
  environments, reads greatly outnumber writes.
  Therefore any factor that benefits read
  performance at the expense of write performance
  will improve server performance most of the time.
  

15

• Secondly, writing to two drives does not cut
  write performance in half. Performance is cut by
  10-20.
• Write request are generated serially, but the
  actual writes are performed in parallel. The
  extra time for the second write has little impact
  on total time.

16
RAID 2

• Distributes data at the bit level
• Uses multiple bits to store parity data and uses
  a large number of individual drives
• Large amount of processing overhead makes it
  unsuitable for database servers.
• RAID 2 is useful for special purpose servers such
  as digital video servers.

17
RAID 3

• Distributes data at the byte level
• Dedicates one drive for storage of parity bits
• In a 4 disk system, the first three disks stores
  data and the fourth handles the parity bits.
• RAID 3 is optimized for special purpose servers
  such as digital video servers and is
  inappropriate for random access areas such as PC
  LANs.

18
(No Transcript)
19
RAID 4

• Similar to RAID 3, except it stripes data at the
  block rather than byte level - providing better
  read performance than RAID 3.
• The small chunk size of RAID 3s means every read
  requires participation from every disk in the
  array. As such RAID 3 is referred to as being
  synchronized or coupled
• RAID 4 larger chunk size means single disk may be
  accessed (unsynchronized or decoupled)
• RAID 4 is not for PC LANs.

20
RAID 5
21

• Most common RAID for PC LANs
• Parity data is striped across all disks taking up
  the equivalent of one disk.
• The equivalent of one drive is unavailable to the
  operating system.
• RAID 5 is optimized for transaction-processing
  activity, in which users frequently read and
  write small amounts of data.

22

• RAID 5 read performance does not need to access
  parity information unless one or more strips are
  unreadable.
• Both data and parity bits are optimized for
  sequential reads.
• Allows parallel read and write, attaining better
  performance on random reads.
• Matches RAID 0 on sequential reads.

23

• RAID 5 write performance writing is more
  problematic.
• Write requires 4 disk operations
• one read for data and another for existing parity
  information. Parity information is recalculated
  based on the read and pending write. Two writes
  are performed (one for data, one for parity
  info).
• This compares to one operation for RAID 0.
• Guard against system failure after data but
  before parity information is written.

24
Stacked RAID
File Data
Sector 1
Sector 2
Sector 4
Sector 3
Sector 5
Sector 6
Sector 7
Sector 8
Sector 10
Sector 9
Array 2
Array 1
Sector 1
Sector 1
Sector 2
Sector 2
Sector 3
Sector 3
Sector 4
Sector 4
Sector 5
Sector 6
Sector 6
Sector 5
Sector 7
Sector 7
Sector 8
Sector 8
Sector 10
Sector 9
Sector 9
Sector 10
Disk 1
Disk 4
Disk 2
Disk 3
25
Stacked RAID

• RAID arrays are seen as a single logical disk by
  the operating system.
• This allows us to stack arrays, using one level
  to control an array of arrays (individual disks
  are replaced by arrays operating at the same or
  different levels)

26

• This allows us to gain the benefits of multiple
  levels without the disadvantages.
• High performance RAID is visible while lower
  performance RAID providing redundancy is hidden
• E.g. RAID 0/1 - combines RAID 0 stripping with
  RAID 1 redundancy.

27
Array Selection (failure)

• RAID 0 failure results in data loss
• RAID 1 - data is read from mirrored drive
• RAID 3 - parity drive failure results in no data
  loss. If data drive fails, reads of other drive
  plus parity drive is used to reconstruct the
  data.
• RAID 5 - failure results in loss of some data and
  some parity information. Reading of other drives
  is used to reconstruct the data.

28
Specifying RAID Implementation

• Decision is made based on
• the type of data to be stored (large vs. small
  writes)
• importance of performance vs. safety of data.
• Cost
• e.g. RAID 0 offers high performance RAID 1
  offers redundancy RAID 3 good performance and
  safety but is poor for PC LANs Stacked RAID
  offers good performance and safety but is
  expensive.