Chapter 6: Storage

1
Chapter 8
Multimedia Storage
2
Magnetic Media

• Magnetic disks are
• Suitable for dynamic data that requires frequent
  changes.
• Good access time and high transfer rate.
• Used for data that must be kept online during
  data capturing and processing.
• Suitable for video-on-demand applications where
  large amounts of time dependent information must
  be transferred at high bit rates.

3
RAID

• Redundant Arrays of Inexpensive Disks.
  (developed at UC Berkeley in 1987)
• Use parallelism between multiple disks to improve
  aggregate I/O performance
• Something like parallelism from multiple CPUs
• Data is distributed across several physical disks
• As an alternative to single large expensive disk
  (SLED) in traditional mainframe systems
• Several levels of RAID, seeking to optimize among
• Performance
• Availability
• Cost

4
RAID (2)

• Advantages
• high data transfer rate for large data accesses.
• high I/O rates (short queuing time) on small data
  accesses.
• uniform load balancing across all of the disks.
• Disadvantages
• Large disk arrays are highly vulnerable to disk
  failures gt Need to add redundancy for better
  availability gt write overhead!

5
Data Striping

• distribute data transparently over multiple disks
  to make them appear as a single, fast, large
  disk.
• multiple I/Os can be served in parallel gt better
  performance
• parallel independent requests gt shorter queuing
  time
• parallel accesses of a single request gt higher
  transfer rate

6
Data Striping (2)

• granularity of data interleaving
• fine distributes the data so that all of the
  arrays disks cooperate in servicing every
  requestgt high I/O transfer rate. Typical
  stripe size 1 bit / 1 byte / 512 bytes.
• 1 bit stripe size speed up every single disk
  access
• 512 bytes stripe size sometimes may not have any
  sped up e.g small disk access of 100 bytes.So
  the disk access time is still bounded by the
  slowest disk in the group.

7
Redundancy

• redundancy is needed to tolerate disk failures
• data and parity distribution
• how redundant information is computed?
• where shall the redundant information reside?
• Hamming error correction
• XOR

8
RAID Level 0 (Nonredundant)

• Data striping only.
• Stripe size segment e.g. 512 bytes
• No data protection redundancy.
• No need to write redundant information gt best
  write performance.
• Read performance ok.
• Any disk failure gt data loss.
• Used in supercomputing environments where
  performance and capacity, rather than
  reliability, are the primary concerns.

9
RAID Level 1 (Disk Mirroring)

• Use twice as many disks as level 0.
• Data is duplicated, called mirroring, or
  shadowing.
• Read is faster, but write is slightly slower
  (why?)
• If a disk fails, its mirror copy can still serve.
• Used in database application where availability
  and transaction rate are more important than
  storage efficiency.

Controller
channel
channel
redundant data (mirror)
10
RAID 1 (2)

• Compare the following 3 RAID-1 configurations.

0
0
2
2
Simple Shadowing
1
1
3
3
0
1
2
3
Declustering
3
0
1
2
0
1
2
3
Chained Declustering
1c
0a
0b
0c
2b
2c
1a
1b
3a
3b
3c
2a
11
RAID Level 2 (Error Correction)

• Uses Hamming code
• Bit interleaving (bit-level data striping)
• For n disks, about log(n) of them store redundant
  data. (More space efficient than mirroring).
• If a disk fails, multiple redundant disks need to
  be read to identify the bad one. However, only
  one redundant disk needs be read to recover the
  lost data.
• No practical use.

12
A Hamming Code Example

• Suppose we want to encode 4 bit information
• We distribute the bit at the following four red
  locationsx1 x2 x3 x4 x5 x6
  x7
• The blue bits are redundant bits for error
  detection and correction.
• Next, we calculate the following 3 equations to
  find x1, x2, x4
• This code can detect and correct 1 bit error

13
A Hamming Code Example (2)

• Example encoding

14
A Hamming Code Example (3)

• Suppose there is 1 bit error at x6
• e.g. original 0 1 0 0 1 0 1 received 0 1 0 0 1
  1 1
• To detect and correct error we calculate
• Notation r is used to indicate it is the current
  data which contains error.
• Obviously, if all b are 0, there is no error,
  otherwise, there is error
• Believe or not, the error must be located at
  position b2b1b0

e.g.
15
RAID Level 3 (Bit-Interleaved Parity)

• Hamming code can detect 1 bit error, but require
  3 redundant bits to tell which bit is wrong.
• However, useless in disk application, because we
  always know which disk fails.
• If 1-bit recovery is needed, simple XOR parity is
  enough.
• Bit-interleaving.

16
RAID Level 4 (Block-Interleaved Parity)

• Note that a single parity disk is enough to
  recover data lost due to single disk failure.
• Block level interleaving
• Small read gt access one data disk large read gt
  access many data disks small write gt 4 I/O
  (read the data disk, compute the difference
  between the old and new images, update the data
  disk, update the parity disk
• Read is fast. How about write?
• If one disk is dedicated for parity, bottleneck
  at parity disk due to writing.
• Easy to implement, high transfer rate.

17
Enhancing RAID-4

• Problems with RAID-4?

18
Distributing Parity

• Parity disk
• simplify the mapping of logical addresses to disk
  addresses.
• every write must update the associated bits on
  the single parity disk. (Fine for fine-grained
  data striping, bad for coarse.)
• Striped parity
• can perform parallel parity update
• Declustered parity
• logically equivalent to combining several smaller
  arrays protected by striped parity into a large
  one.

19
RAID Level 5(Block-Interleaved
Distributed-Parity)

• Eliminates the parity disk bottleneck by
  distributing the parity uniformly over all of the
  disks.
• Improves read performance by allowing all disks
  be used to serve read requests.
• Best for small reads, large reads, large writes.

20
RAID Level 6 (PQ Redundancy)

• Uses 2 redundant disks to protect up to two disk
  failures.
• Compute 2 different parities instead of 1.
• Similar read performance as with Level 5, but
  write is slightly worse

21
Optical Media

• Well accepted because
• High storage capacity
• Random access to data
• Life span of more than 30 years (c.f. ltlt 20 years
  for magnetic media)
• Removable and portable

22
History of Optical Media

• Optical videodisk was invented by Friebus in
  1929. Prototype using laser to record and read
  was demonstrated by Phillips and MCA in 1972.
• Videodisks developed by Philips has been
  commercially available since 1978.
• Then compact disk technology for digital audio
  (CD-DA) came out in early 1980s.
• The use of optical disks for digital data storage
  came with the introduction and improvement of
  CD-ROM during the 1980s.

23
Optical Disk Technology

• Optical storage media use the intensity of
  reflected laser light as an information source.

24
Optical Disk Technology (2)

• An optical disk consists of 3 layers
• Protective layer (only 0.002mm thick on the label
  side).
• Reflective layer (aluminum coating).
• Substrate layer (transparent).

• In the factory, depressions are cut on the disk
  surface, forming lands and pits (0.12um
  different in heights).

25
Optical Disk Technology (3)

• Simple thresholding yields the H and L readback.

• Do you know
• that data are read from the disk inside-out?
• that a CD should be cleaned radially?

26
Advantages of Optical Media

• Continuous data stream. Data stored in spiral or
  concentric tracks. For the spiral track storage,
  data can be easily played back in a continuous
  data stream.
• High density. Distance between tracks is 1.6um,
  each track is 0.6um wide, i.e., 1 bit per sq.um
  or 1Mb per sq.mm. Floppy disk has 96 tracks per
  inch, optical disk has 16000 tracks per inch.
• Long life. Magnetization can decrease over time.
  Lands and pits not changed unless physically
  damaged.
• Low wearing. Laser source in head can be
  positioned at 1mm from disk surface. Does not
  have to be as close to the surface as with
  magnetic disks. It reduces friction and increases
  life span.

27
Digital Optical Disks

• Audio CD was developed by Philips and Sony in
  1982.
• Basic technology extended to 550 MB CD-ROM in
  1985.
• When used for multimedia, storage capacity is
  inadequate for motion video, and data rate
  limited to 1.5Mbps.
• CD-ROM/XA and CD-I announced in 1986 and 87 to
  support applications of text, images, audio and
  FSFM video.
• Recent developments include WORM (write once read
  many), MO (magneto optical), CD Recordable disks,
  and DVD.

28
Digital Optical Disk (2)

• Why CD is slower than hard disk?
• CD is originally designed for squeezing as much
  music data into the disk as possible. The density
  of data is same in inner and outer tracks.gt The
  disk has to rotate slower when reading the outer
  trackgt Variable speed is slow to adjust for
  random access (as in computer-based multimedia
  application)
• Optical disk head is heavier than magnetic heads.
  More inertia takes longer seek time for head
  movements.

29
Uses of Optical Disks
30
CD-DA(Compact Disk Digital Audio)

• 1982 by Philips and Sony.
• 12cm diameter, 1.2 mm thick optical disk,
  stores/plays in CLV. Spiral tracks of about
  20,000 windings in total.
• Data are recorded such that pit-to-land and
  land-to-pit transitions are coding 1s. 0s are
  coded as no transition.
• Pits and lands are not directly used to represent
  digital information. How can you represent 11?
• Redundancy added to break up consecutive 1s and
  0s.

31
CD-DA

• Data rate 44.1KHz sampling, 16-bit quantization,
  175KBytes/sec.
• Capacity 747MB, up to 74 min high-quality sound.
• Capability of random access to tracks and index
  points.
• Error rate as low as 10(-8). However, still
  not low enough for computer data.

32
8 to 14 Modulation (EFM)

• Pits and lands may not follow too closely one
  after another on a CD-DA. Rule 1 between any 2
  1s, there are at least 2 0s.
• For synchronization, pit or land sequences are
  not allowed to be too long. Rule 2 at most 10
  0s can follow one after another.
• Solution Map every 8 bit pattern into a 14 bit
  pattern that satisfies the 2 rules. Among the
  214 patterns, 267 of them are valid gt just
  fit.Also, between consecutive 14-bit sequences,
  3 merging bits are added to enforce the rules.

33
8 to 14 Modulation (Example)
34
Low Level Data Encoding

• Thus, an eight-bit byte of actual data is encoded
  into a total of 17 channel bits.
• For synchronization and error correction, every
  24 bytes of data is packaged into a frame
• sync pattern (24 3 bits)
• control byte (17 bits)
• 12 data bytes (12 17 bits)
• 4 error correction bytes (4 17 bits)
• 12 data bytes (12 17 bits)
• 4 error correction bytes (4 17 bits)
• Total 588 channel bits for 192 actual data bits.

35
First Level Error Correction

• Cross Interleave Reed-Solomon Coding.
• Recall that each frame contains 24 data bytes and
  8 error correction bytes.
• The first 4 correction bytes cover the frames
  data. The other 4 correction bytes cover data
  over 7 frames.
• When a frame is read, the first 4 correction
  bytes are checked. If not ok, the decoder decodes
  the data bytes after subsequent correction codes
  are read.
• 7 frames 7.7 mm track length. Try radially
  scratch your CD with a cutter and see if it still
  works.

36
CD-ROM (Compact Disk Read Only)

• 1985 by Philips and Sony.
• Tracks are divided into audio and data types.
  Disk containing both types are called Mixed Mode
  Disk.
• It operates in 2 modes mode 1 is for computer
  data, and mode 2 is for media data.
• Mode 1
• Error rate requires better than 10(-8) for
  computer data. Mode 1 achieves 10(-12) error
  rate by using a second level error correction..

37
CD-ROM (2)

• Random access to subtrack units called blocks
  (2352 bytes). (For CD-DA, random access is on
  track level only.)
• Mode 1 for computer data. A capacity of 333,000
  blocks to be played in 74 min, i.e. 660MB storage
  with data rate of 150KBps. Each block consists of
  32 frames (_at_588 bits each).
• Mode 2
• Mode 2 holds data of any media.
• Additional error correction not crucial, so not
  used.
• Disk has capacity of 750MB and a data rate of
  175KBps.

38
CD-ROM (3)

• CD-ROM is a very economic medium for publication
  and distribution.
• Limitations of CD-ROM
• Random access to a CD track can be anywhere from
  200ms up to 1 sec in access time.
• Continuous media stored sequentially in CD-ROM
  tracks. Although important for multimedia
  applications, simultaneous playback of audio and
  other data is not possible.

39
CD-ROM/XA (Extended Architecture)

• 1989, established by Microsoft, Philips and Sony.
• Based on CD-ROM and CD-I.
• Goal concurrent output of several media. Within
  1 track, blocks of different media can be stored.
  It allows interleaved storage and retrieval of
  multimedia data.
• A sub-header is added to each block to describe
  the block.
• CD-ROM/XA uses CD-ROM mode 2 to define actual
  blocks. Two forms

40
CD-ROM/XA (2)

• Form 1 provides more error detection/correction
  at the expense of redundancy. 2048 bytes (of
  2352) are for user data. Form 2 allows 13
  more storage for user data, but at the expense of
  the error correction.

41
CD-R (Compact Disk Recordable)

• CD-R allows tracks to be recorded once.
• 4 layers protective, reflective, absorption, and
  substrate.

Traditional CD-ROM
CD-R Media
Lacquer
Lacquer
Gold
Dont leave out in sunlight
Aluminum
Dye
Polycarbonate
Polycarbonate
Molded by stamper
Burned by high power laser beam
42
CD-R(2)

• Land and pit reflections realized by irreversible
  thermal effect (above 250C) on the absorption
  layer.
• Playable on CD players.

43
CD-R (3)

• Recording sessions
• A CD has 3 areas lead-in, actual data, lead-out.
• Lead-in includes the table of contents
  directory, indices to individual tracks.
• Data area include all tracks where actual data is
  stored.
• Lead-out marks the end of the data area.
• Multiple sessions of lead-in, data, lead-out can
  be written separately over time.
• During 1 write activity, all data for a session
  are written with their table of contents, after
  which the session can be played on any CD player.

44
CD-MO(Compact Disk Magneto Optical)

• Specification published by Philips and Sony in
  1991.
• Working principle is different from other CD
  technologies. (Incompatible with other CD
  formats.)
• Based on the polarization of light by magnetic
  field.
• Disk surface is light reflecting magnetic
  substrate.
• During writing, surface is heated to above 150C,
  and magnetic field polarizes individual dipoles.
• During reading, surface is irradiated with a
  laser beam, polarization of laser light changed
  according to the magnetization.

45
Digital Versatile Disk (DVD)

• Also called Digital Video Disk.
• Capacity 4.7 to 17 GB (25 CDs).
• Q Is it a good idea to replace VHS tapes by DVD
  disks in video rental stores?
• Digital video can be stored and distributed more
  cheaply, also it allows interactivity.
• Can be used to store up to 133 minutes (8-9 hrs
  for high capacity ones) of studio quality video
  and multi-channel surround-sound audio, or 30
  hours of CD-quality audio.

46
DVD (2)

• DVD achieves a greater capacity by
• minimum pit length is reduced from 0.834 micron
  (CD) to 0.4 micron (DVD).
• inter-track space is reduced from 1.6 micron (CD)
  to 0.74 micron (DVD).

47
DVD (3)

• To read the condensed pits, DVD uses a laser of
  shorter wavelength (635-650 nm for CD it is 780
  nm).
• Reducing the pit size and track distance
  increases the discs capacity to 4.7GB.
• Dual layering. A semireflective layer (3.8GB) on
  top of a fully reflective layer (4.7GB) gt 8.5GB
  total.
• Double side. Two substrates bonded back-to-back.
  Each side could have one layer or two layers gt
  capacity ranges from 9.4GB to 17GB.

48
Some DVD drives can also read CDs