Download Storage Technology

Chapter Goals
• Describe the distinguishing characteristics of
primary and secondary storage
• Describe the devices used to implement primary
storage
• Describe memory allocation schemes
• Compare and contrast secondary storage
technology alternatives
Systems Architecture, Fifth Edition
2
Chapter Goals (continued)
• Describe factors that determine storage device
performance
• Choose appropriate secondary storage
technologies and devices
Systems Architecture, Fifth Edition
3
Systems Architecture, Fifth Edition
4
Systems Architecture, Fifth Edition
5
Characteristics of Storage Devices
•
•
•
•
•
•
Speed
Volatility
Access method
Portability
Cost and capacity
Let’s examine each of these...
Systems Architecture, Fifth Edition
6
Speed
• Primary storage speed
– Typically faster than secondary storage speed by a
factor of 105 or more
– Expressed in nanoseconds (billionths of a second)
• Secondary storage speed
– Expressed in milliseconds (thousandths of a second)
• Data transfer rate = 1 second/access time (in seconds)
x unit of data transfer (in bytes) (simplified)
Systems Architecture, Fifth Edition
7
Volatility
• Primary storage devices are generally volatile
– Cannot reliably hold data for long periods
• Secondary storage devices are generally
nonvolatile
– Hold data without loss over long periods of time
Systems Architecture, Fifth Edition
8
Access Method
• Serial access (linear, such as a tape)
• Random access (direct access, such as RAM)
• Parallel access (simultaneous access, such as
RAID)
Systems Architecture, Fifth Edition
9
Portability
• Removable storage media with standardized
formats (e.g., compact disc and tape storage)
• Typically results in slower access speeds
Systems Architecture, Fifth Edition
10
Cost and Capacity
• Cost increases:
– With improved speed, volatility, or portability
– As access method moves from serial to random to
parallel access method
• Primary storage - expensive (high speed and
combination of parallel/random access methods)
• Capacity of secondary storage devices is greater
than primary storage devices
Systems Architecture, Fifth Edition
11
Systems Architecture, Fifth Edition
12
Memory-Storage Hierarchy
Systems Architecture, Fifth Edition
13
Primary Storage Devices
• Critical performance characteristics
– Access speed
– Data transfer unit size
• Must closely match CPU speed and word size to
avoid wait states
Systems Architecture, Fifth Edition
14
Storing Electrical Signals
• Directly
– By devices such as batteries and capacitors
– Trade off between access speed and volatility
• Indirectly
– Uses energy to alter the state of a device; inverse
process regenerates equivalent electrical signal
• Modern computers use memory implemented with
semiconductors (RAM and NVM)
nonvolatile memory
Systems Architecture, Fifth Edition
15
Random Access Memory
• Characteristics
– Microchip implementation using semiconductors
– Ability to read and write with equal speed
– Random access to stored bytes, words, or larger
data units
• Basic types
– Static RAM (SRAM) – faster than DRAM; uses 6
transistors/bit; does not need refreshing; faster than DRAM
– Dynamic RAM (DRAM) – uses 1 transistor and 1
capacitor / bit; cheaper than SRAM; needs refreshing; higher
density; needs more power requirements
Systems Architecture, Fifth Edition
16
Random Access Memory
• To bridge performance gap between memory and
microprocessors
– Read-ahead memory access
– Synchronous read operations
– On-chip memory caches
Systems Architecture, Fifth Edition
17
Nonvolatile Memory
• Random access memory with long-term or
permanent data retention
• Usually relegated to specialized roles and
secondary storage; slower write speeds and limited
number of rewrites
• Generations of devices (ROM, EPROM, and
EEPROM)
Systems Architecture, Fifth Edition
18
Nonvolatile Memory
• Flash RAM (most common NVM)
– Competitive with DRAM in capacity and read
performance
– Relatively slow write speed
– Limited number of write cycles (more on this later)
• NVM technologies under development
– Ferroelectric RAM
– Polymer memory
Systems Architecture, Fifth Edition
19
Memory Packaging
• Dual in-line packages (DIPs)
– Early RAM and ROM circuits
• Single in-line memory module (SIMM)
– Standard RAM package in late 1980s
• Double in-line memory module (DIMM)
– Newer packaging standard
– A SIMM with independent electrical contacts on
both sides of the module
Systems Architecture, Fifth Edition
20
Systems Architecture, Fifth Edition
21
CPU Memory Access
• Critical design issues for primary storage devices
and processors
– Physical organization of memory
– Organization of programs and data within memory
– Method(s) of referencing specific memory
locations
Systems Architecture, Fifth Edition
22
Physical Memory Organization
• Physical memory
– Actual number of memory bytes that physically are
installed in the machine
• Most and least significant bytes
• Big endian (stores most significant byte/bit at
lowest memory address) and little endian
• Addressable memory
– Highest numbered storage byte that can be
represented
Systems Architecture, Fifth Edition
23
Systems Architecture, Fifth Edition
24
Systems Architecture, Fifth Edition
25
Memory Access Time
• Memory rated at PCxxxx delivers peak bandwidth
of xxxxMB/sec.
• For example, PC3200 memory delivers 3.2GB/sec
peak bandwidth.
• Typically, there is a significant delay before you
can get back the first byte (say 6 cycles)
• Random accesses would deliver around
500KB/sec.
Systems Architecture, Fifth Edition
26
Memory compared to CPU
• How much data can a CPU pump around?
• 2GHz = 2,000,000 cycles per second
• Each cycle, the CPU can move around a word
(4 bytes on a 32-bit machine)
• so 8GB/sec
Systems Architecture, Fifth Edition
27
Magnetic Storage
• Exploits duality of magnetism and electricity
– Converts electrical signals into magnetic charges
– Captures magnetic charge on a storage medium
– Later regenerates electrical current from stored
magnetic charge
• Polarity of magnetic charge represents bit values
zero and one
Systems Architecture, Fifth Edition
28
Systems Architecture, Fifth Edition
29
Systems Architecture, Fifth Edition
30
Systems Architecture, Fifth Edition
31
Magnetic Tape
• Ribbon of plastic with a coercible (usually
metallic oxide) surface coating
• Mounts in a tape drive for reading and writing
• Relatively slow serial access
• Compounds magnetic leakage; wraps upon itself
• Susceptible to stretching, friction, temperature
variations
Systems Architecture, Fifth Edition
32
Systems Architecture, Fifth Edition
33
Magnetic Tape
• Two approaches to recording data
– Linear recording
– Helical scanning
• Several formats and standards (e.g., DDS [DAT],
AIT, Mammoth, DLT (digital linear tape), LTO
(linear tape-open))
• Highest capacity tapes hold about 1 TB (same as
largest disk drives)
• But much cheaper per byte
Systems Architecture, Fifth Edition
34
Systems Architecture, Fifth Edition
35
Modern Tape Formats
and Capacities (uncompressed)
•
•
•
•
•
DDS (DAT), from Sony and HP: 2-36 GB
AIT, from Sony: 35-400 GB
Mammoth, from Exabyte: 20-80 GB
DLT, from Quantum: 20-600 GB
LTO, from HP, IBM, Seagate: 100-800 GB
Systems Architecture, Fifth Edition
36
Magnetic Disk
• Flat, circular platter with metallic coating that is
rotated beneath read/write heads
• Random access device; read/write head can be
moved to any location on the platter
• Hard disks and floppy disks
• Cost performance leader for general-purpose
on-line secondary storage
Systems Architecture, Fifth Edition
37
Systems Architecture, Fifth Edition
38
Systems Architecture, Fifth Edition
39
Magnetic Disk Access Time
•
•
•
•
Head-to-head switching time
Track-to-track seek time
Rotational delay
Most important performance numbers
– Average access time
– Sequential access time
– Sustained data transfer rate
Systems Architecture, Fifth Edition
40
Average Access Time
• Head switching time (negligible)
• Plus head seek time (given in ms)
• Plus rotational delay (on average, ½ a turn)
If disk spins at 6000RPM, what is the rotational
delay?
– One turn takes 1/6000 min or 1/100 sec = 10ms
– ½ turn takes 5ms.
Systems Architecture, Fifth Edition
41
Average Access Time
• Plus read time (time to spin an entire sector)
– If the drive spins at 6000RPM and the disk has 20
sectors per track, what is the read time?
– Time for 1 full spin is
– Time for 1/20 of a spin is
Systems Architecture, Fifth Edition
1
1
min 
sec  10ms
6000
100
10ms 
1
 0.5ms
20
42
Let’s Try Another Problem
• Drive spins at 7200RPM and has average seek
time of 8ms. The disk has 24 sectors per track.
What is the average access time?
• Head seek time = 0.008 sec (given)
• Rotational delay (1/2 spin) = 7200 RPM (1/2) =
120 RPS(1/2) = 1/120 sec/rev(1/2) = 0.0042 sec
• Read time (1 sector) = 0.0084 (full spin) / 24 sectors per track =
0.00035 sec
• Total = 0.008 + 0.0042 + 0.00035 = 0.01255 sec or 12.55 ms
• This is the average access time
Systems Architecture, Fifth Edition
43
Sequential Access Time
• The best possible read time is if the head is in
exactly the right place (like when we read
consecutive sectors)
• Sequential access time is the amount of time for
one sector to spin under the head
• Drive spins at 7200RPM and has average seek
time of 8ms. The disk has 24 sectors per track.
What is the sequential access time?
Systems Architecture, Fifth Edition
44
Sequential Access Time
• Drive spins at 7200RPM and has average seek
time of 8ms. The disk has 24 sectors per track.
What is the sequential access time?
Read time (1 sector) = 0.0084 (full spin) / 24 sectors
per track = 0.00035 sec
(no need to include head seek time and rotational delay)
Systems Architecture, Fifth Edition
45
Data Transfer Rate
• The data transfer rate is the number of access that
can be made per second times the amount of data
per transfer.
• The size of a transfer is the size of a sector,
typically 512 bytes.
Systems Architecture, Fifth Edition
46
Maximum Data Transfer Rate
• Maximum data transfer rate is computed using the
best possible access time (sequential access time).
• In the prior example, sequential access time was
0.35ms = 0.00035 seconds
• Accesses per second is
1
0.00035
 2857.1
• Transfer rate is
2857.1 512 bytes  1,462,835 bytes  1.4MB/sec
Systems Architecture, Fifth Edition
47
Sustained Data Transfer Rate
• Sustained data transfer rate is computed using the
average access time (sectors may not be contiguous).
• In the prior example, average access time was
12.55ms = 0.01255 seconds
• Accesses per second is
1
0.01255
 79.68
• Transfer rate is
79.68  512 bytes  40,796.2 bytes  40KB/sec
Systems Architecture, Fifth Edition
48
Systems Architecture, Fifth Edition
49
To increase capacity per platter, disk manufacturers
divide tracks into zones and vary the sectors per track
in each zone.
Systems Architecture, Fifth Edition
50
Optical Mass Storage Devices
• Store bit values as variations in light reflection
• Higher areal density and longer data life than
magnetic storage
• Standardized and relatively inexpensive
• Uses: read-only storage with low performance
requirements, applications with high capacity
requirements, and where portability in a
standardized format is needed
Systems Architecture, Fifth Edition
51
Optical storage devices read data by shining laser
beam on the disc.
Systems Architecture, Fifth Edition
52
Systems Architecture, Fifth Edition
53
CD-ROM
• Read-only; data permanently embedded in durable
polycarbonate disc
• Bit values represented as flat areas (lands) and
concave dents (pits) in the reflective layer
• Data recorded in single continuous track that
spirals outward from center of disc
• Popular medium for distributing software and
large data sets
Variable data transfer rate
Systems Architecture, Fifth Edition
54
CD-ROM
Advantages
Drawbacks
• Standardized format
• High density
• Cheap to manufacture
• Cannot be rewritten
• Capacity limited to 700
MB
Systems Architecture, Fifth Edition
55
CD-R
• Uses a laser that can be switched between high
and low power and a laser-sensitive dye embedded
in the disc
• Relatively cheap
• Common uses: create music CDs on home
computers, back up data from other storage
devices, create archives of large data sets, and
manufacture small quantities of identical CDs
Systems Architecture, Fifth Edition
56
Phase-Change Optical Discs
• Enables nondestructive writing to optical storage
media
• Materials change state easily from non-crystalline
(amorphous), to crystalline, and then back again
– Reflective layer is a compound of tellurium,
selenium, and tin
• Example: CD-RW
Systems Architecture, Fifth Edition
57
DVD
• Improves on CD and CD-RW technology
– Increased track and bit density: smaller wavelength
lasers and more precise mechanical control
– Improved error correction
– Multiple recording sites and layers
Systems Architecture, Fifth Edition
58
Flash Drives
• AKA USB flash drives, or UFD
• NAND-type flash memory integrated with a USB
connector
• Replacing floppy disk/diskette
• Lightweight and compact, minimal model contains
a printed circuit board with storage controller
device, flash memory chip, crystal oscillator, and
LED
Systems Architecture, Fifth Edition
59
Flash Drive Internals
From Wikipedia
Systems Architecture, Fifth Edition
60
Flash Drive Technology
• Based on earlier EPROM technology (UVerasable)
• EEPROMs (electrically erasable) replaced
EPROMs
• To update contents, an entire “region” of memory
has to be moved off the flash drive, updated, the
flash drive contents are erased, and the data
moved back to the flash drive
Systems Architecture, Fifth Edition
61
Common Uses
• Very popular with network administrators as they
load them with configuration information as well
as software used for maintenance and recovery
• Operating system boots
• Application carriers
• MP3 music players
• Windows Vista ReadyBoot
• And of course file backup
Systems Architecture, Fifth Edition
62
Major Weaknesses
• Limited number of erase and write cycles
• Average device should support several hundred
thousand cycles, but write operations will slow
with age of device
• Not as fast as fixed disk drives
Systems Architecture, Fifth Edition
63
RAID
• Redundant Array of Independent Drives, is a
collection of techniques to interface multiple hard
disk drives to a computer
• Mostly, they are a collection of techniques to store
data redundantly on multiple hard disk drives
• Only the first RAID technique, RAID 0, does not
store the data redundantly
Systems Architecture, Fifth Edition
64
More Common RAID
Techniques – RAID 0
• RAID 0, in which the data is broken into pieces
and each piece is stored on different disk drives.
There is no redundancy of data in this technique,
so if one disk drive fails, some of the data is lost.
The advantage of this technique is the speed in
which data can be read or written across multiple
disks at the same time.
• aka striped
Systems Architecture, Fifth Edition
65
RAID 0
From Wikipedia
Systems Architecture, Fifth Edition
66
RAID 0
• When two drives are used and they are of different
sizes, use the smaller size and double
• For example, one drive is 300 GB and the other is
200 GB, then you have a total of 400 GB of
storage
• RAID 0 useful in large, NSF servers (mounting
issues) and in Microsoft OS where you might run
out of letter designations for drives
Systems Architecture, Fifth Edition
67
RAID 1
• RAID 1, in which the data is stored on at least two
disk drives in duplicate so as to provide a level of
redundancy (fault tolerance) should one disk
become corrupted
• This technique is also known as disk mirroring, or
mirrored
Systems Architecture, Fifth Edition
68
RAID 1
From Wikipedia
Systems Architecture, Fifth Edition
69
RAID 4
• RAID 4, in which the data is striped across
multiple disk drives (based on blocks) and error
checking information (parity checks) concerning
the stored data is kept along with the data (but on
a separate disk).
• This error checking information can be used to
detect errors and possibly reconstruct the data
should some of it become corrupted.
Systems Architecture, Fifth Edition
70
RAID 4
From Wikipedia
Systems Architecture, Fifth Edition
71
RAID 5
• Employs block-level striping with parity
information distributed across all data disks
Systems Architecture, Fifth Edition
72
Other RAIDs
• RAID 2 – stripes data at the bit level and uses a
Hamming code (no implementations)
• RAID 3 – stripes data at the byte level and uses
parity checking
• RAID 6 – block-level striping with two parity
blocks spread over all data disks
• And multi-level RAIDs, such as RAID 0-5
Systems Architecture, Fifth Edition
73
Summary
• Storage devices and their underlying technologies
• Characteristics common to all storage devices
• Technology, strengths, and weaknesses of primary
and secondary storage
Systems Architecture, Fifth Edition
74