External Memory Patrick Happ Raul Queiroz Feitosa Parts of these - - PowerPoint PPT Presentation

External Memory

Patrick Happ Raul Queiroz Feitosa

Parts of these slides are from the support material provided by W. Stallings

Objective

“This chapter examines a range of external memory devices and systems.”

• W. Stallings

Outline

 Magnetic Disc  RAID  Solid State Drives  Optical Memory  Magnetic Tape

Magnetic Disc

 Disc substrate of non magnetic material coated

with magnetizable material (iron oxide…rust)

 Substrate used to be aluminium; now glass

 Improved surface uniformity→ Increases reliability  Reduction in surface defects →Reduced read/write errors  Lower flight heights (See later)  Better stiffness  Better shock/damage resistance

Read and Write Mechanisms

Underlying Physics

Current flowing through a conducting coil creates a magnetic field that orients the magnetic domains over the metal Changes in the magnetic field intensity induces a current in a coil

Write Operation



The electronic in the drive receives binary data and converts it into a current that flows through the coil.



The current flow direction changes at each “1” and keeps unchanged at each “0”



The interaction with the media magnetizes the material, whose direction depends on the current direction in the coil.

Read Operation (traditional)

 The same coil for read

and write

 Magnetic field variation

due to the movement relative to coil produces current.

 The direction of the

induced current indicates what is recorded.

Read Operation (contemporary)



Separate read head, close to write head



Partially shielded magneto resistive (MR) sensor



Electrical resistance depends on direction of magnetic field



High frequency operation

Higher storage density and speed

Inductive Write MR Read

Disc Data Layout

Concentric rings or tracks

 Gaps between tracks  Reduce gap to increase

capacity

 Same number of bits per

track (variable packing density)

 Constant angular velocity

Tracks divided into sectors

Disc Velocity

 Bit near centre of rotating disc passes fixed point slower

than bit on outside of disc.

 Increase spacing between bits in external tracks.  Disc rotates at constant angular velocity (CAV)

 Gives pie shaped sectors and concentric tracks  Individual tracks and sectors addressable  Move head to given track and wait for given sector  Waste of space on outer tracks → Lower data density

 Can use zones to increase capacity

 Within a zone (typically 16) bits per track is constant  Zones farther/closer to the centre contain more/less sectors.  More complex circuitry

Disc Layout Methods Diagram

Characteristics

 Fixed (rare) or movable head  Removable or fixed  Single or double (usually) sided  Single or multiple platter  Head mechanism

 Contact (Floppy)  Fixed gap  Flying (Winchester)

Fixed/Movable Head Disc

 Fixed head

One read write head per track Heads mounted on fixed ridged arm

 Movable head

One read write head per side Mounted on a movable arm

Removable or Not

 Removable disc

 Can be removed from drive and

replaced with another disc

 Provides unlimited storage

capacity

 Easy data transfer between

systems

 Non-removable disc

 Permanently mounted in the

drive

Head Mechanism

 Contact (Floppy)  Fixed gap  Flying (Winchester)

Conventional Hard Disc

Hard Disc 3D Visualization

Inside the Hard Disc

Cylinders through Multiple Platters

A Portion of a Disc Track.

Two sectors

Winchester Disc Format Seagate ST506

Disc Controller

Typically embedded in the disc drive, which acts as an interface between the CPU and the disc hardware. The controller has an internal cache (typically a number of MBs) that it uses to buffer data for read/write requests.

Speed

 Seek time

 Time to move head to correct track

 (Rotational) latency

 Time it takes for the disc to rotate so that the desired sector

is under the read/write head

 Transfer Time

 Once the read/write head is positioned over the data, this is

the time it takes for transferring data

 Access time

 Seek + Latency (according to Stalling)  Seek + Latency + Transfer (according to Parhami) 

Speed

Exercise 1:

Given average seek time = 4 ms, rotation speed =15,000 rpm, 512 bytes/sector, 500 sectors/track, 5 tracks per cylinder. What is the time to read a file consisting of 2500 sectors for a total

• f 1.28 Mbyte?

Speed

Exercise 1 - Solution 1

We assume that the file is stored as compactly as possible. That is, the file occupies 5 tracks of one cylinder (sequential organization). 15,000 rpm → time for a complete rotation = 60/15000  4ms Transfer → data to transfer / data from track * rotation 500 × 512/ (500 × 512) *4ms = 4ms (Full track) (avg. seek) (avg. rotational latency) (transfer) To read the first track → 4 + 2 + 4 = 10 ms We assume that the tracks are aligned across the cylinders and the time to switch between tracks of the same cylinder is close to zero. (transfer) Time to read the other four tracks → 4 × 4 = 16 ms Time to read the file = 26 ms

Speed

Exercise 1 - Solution 2

We assume random access rather than sequential access. That is, the accesses are distributed randomly over the disc. For each sector we have. 15,000 rpm → time for a complete rotation = 4ms Transfer → data to transfer / data from track * rotation 512/ (500 × 512) *4ms = 4/500 ms (seek) (rotational latency) (transfer) To read the first sector → 4 + 2 + 4/500 = 6.008 ms . Time to read the file = 2500 × 6.008 = 15,020 ms = 15.02 seconds! Fragmentation!

Defragmentation

Speed

Exercise 2:

A hard disc has 500 cylinders, 5 tracks/cylinder, 100 sectors per track and operates at 3000 rpm. The time to move from the most external to the most internal cylinder is equal 10

• ms. Assume that the time to switch between tracks of the

same cylinder is negligible, and disregard acceleration time when the heads move. Compute the average time to:

a)

Read a single sector,

b)

Read the whole first cylinder starting with the first track and going track by track till the last track

c)

Read the whole disc starting with the first cylinder and first track, going track by track , cylinder by cylinder till the final cylinder and track

Speed

Exercise 3: A hard disc has 600 cylinders, 6 tracks/cylinder, 60 sectors per track and operates at 12000 rpm. The time to move from the most external to the most internal cylinder is equal 24

• ms. Assume that the time to switch between tracks of the

same cylinder is negligible, and disregard acceleration time when the heads move. How long does it take to read a file stored in 10 sectors, assuming that

a)

The file is stored in 10 adjacent sectors of the same cylinder and track.

b)

The file is stored in 5 adjacent sectors of the same track of the first cylinder and then in 5 adjacent sectors of the same track of the last cylinder.

c)

The file is stored in 5 adjacent sectors of one track of the first cylinder and then in 5 adjacent sectors of another track of the same cylinder.

Outline

 Magnetic Disc  RAID  Solid State Drives  Optical Memory  Magnetic Tape

RAID - what’s in a name?

 Redundant Array of Independent/ Inexpensive Discs  Set of physical discs viewed as single logical drive by

O/S

The two keywords are:

 Redundant

Redundant data on multiple discs provides fault tolerance

 Array.

An array of multiple discs accessed in parallel will give greater throughput than a single disc.

Non-Redundant - RAID 0

 Data striped across all discs  Round Robin striping  Increase speed

 Multiple data requests probably not on same disc  discs seek in parallel  A set of data is likely to be striped across multiple discs

 No redundancy

strip 8 strip 4 strip 0 strip 9 strip 5 strip 1 strip 10 strip 6 strip 2 strip 11 strip 7 strip 3

Mirrored - RAID 1

 Redundancy is achieved by duplicating all data  A read request can be serviced by either of the two

discs → performance dictated by the fastest one

 A write request requires that both discs be updated →

performance dictated by the slowest one

 Simple recovery – if driver fails data is available in

the second one

mirrored

RAID 1+0

 Data is striped across discs  2 copies of each strip on separate discs (mirroring)  Positive aspects:

 Read from either (the one with least seek)  Write to both – no write penalty (see later)  Recovery is simple - Swap faulty disc & re-mirror (no down time)

 Negative aspect:

 expensive

strip 8 strip 4 strip 0 strip 9 strip 5 strip 1 strip 10 strip 6 strip 2 strip 11 strip 7 strip 3 strip 8 strip 4 strip 0 strip 9 strip 5 strip 1 strip 10 strip 6 strip 2 strip 11 strip 7 strip 3

Memory Style - RAID 2

 Typically, discs are synchronized – head in same position  Very small stripes - often single byte/word  Error correction calculated across corresponding bits.  On a single write, all data on parity discs must be accessed.  Lots of redundancy

 Expensive  Only effective if many disc errors occur → not used nowadays

b0 b1 b2 b3

parity parity parity

Bit-Interleaved Parity - RAID 3

 Similar to RAID 2  Only one redundant disc, no matter how large the array  Simple parity bit for each set of corresponding bits  Data on failed drive can be reconstructed from surviving data

and parity info

 Very high transfer rates bit 1 bit 2 bit 3 bit 4

parity

Block-Interleaved Parity - RAID 4

 Each disc operates independently  separate I/O requests can

be satisfied in parallel.

 Good for high I/O request rate  Large stripes  Bit by bit parity calculated across stripes on each disc  Parity stored on parity disc  Writes involve 2 reads and writes - user and parity stripes.  Parity disc becomes a bottleneck and overloaded.

strip 0 strip 1 strip 2 strip 3

strip 4 strip 5 strip 6 strip 7

strip 8 strip 9 strip 10 strip 11

Block-Interleaved Distributed-Parity - RAID 5

 Like RAID 4  Parity striped across all discs  Round robin allocation for parity stripes  Avoids RAID 4 bottleneck at parity disc  Commonly used in network servers

strip 0 strip 1 strip 2 strip 3 P0-3 strip 4 strip 5 strip 6 P4-7 strip 7 strip 8 strip 9 P8-11 strip 10 strip 11 strip 12 P12-15 strip 13 strip 14 strip 15 P16-19 strip 16 strip 17 strip 18

P+Q redundancy - RAID 6

 Two parity calculations  Stored in separate blocks on different discs  User requirement of N discs needs N+2  High data availability

 Three discs need to fail for data loss  Significant write penalty (30% compared with RAID 5)

Q16-19 strip 1 strip 5 strip 9 P12-15 strip 18 Q4-7 P0-3 strip 10 strip 14 strip 0 strip 4 strip 8 strip 12 P16-19 Q12-15 strip 2 strip 6 P8-11 strip 16 strip 3 P4-7 Q8-11 strip 13 strip 17 Q0-3 strip 7 strip 11 strip 15

RAID - A short Break

 Raid Kills Bugs Dead

RAID Comparison

Outline

 Magnetic Disc  RAID  Solid State Drives  Optical Memory  Magnetic Tape

Solid State Drives

• One of the most significant developments in computer

architecture in recent years!

• Generally complement or even replace hard disk drives

(HDDs).

• Solid state refers to electronic circuitry built with

semiconductors.

• Uses a type of semiconductor memory referred to as flash

memory.

Flash Memory

• Has been around for years and is used in many consumer

electronic products: smart phones, GPS devices, MP3 players, digital cameras, and USB devices.

• The cost and performance has evolved to the point where it is

feasible to use flash memory drives (even to replace HDDs).

Flash Memory

• Transistors exploit the properties of semiconductors: a

small voltage applied to the gate can be used to control the flow of a large current between source and drain.

• A floating gate (insulated by a thin oxide layer) does

not interfere with the transistor when the state is 1.

• Applying a large voltage across the oxide layer causes

electrons to tunnel through it and become trapped on the floating gate, where they remain even if the power is disconnected. State is 0.

• The state of the cell can be read by using external

circuitry to test whether the transistor is working or not.

SSD X HDD

• High-performance input/output operations per second (IOPS):

Significantly increases performance I/O subsystems.

• Durability: Less susceptible to physical shock and vibration.
• Longer lifespan: SSDs are not susceptible to mechanical wear.
• Lower power consumption: SSDs use as little as 2.1 watts of power per

drive, considerably less than comparable-size HDDs.

• Quieter and cooler running capabilities: Less floor space required, lower

energy costs, and a greener enterprise.

• Lower access times and latency rates: Over 10 times faster than the

spinning disks in an HDD.

• HDDs still has a cost per bit advantage and a capacity advantage.

SSD X HDD

Flash Drives Seagate Laptop Internal HDD Copy/write speed 200-550 Mbps 50–120 Mbps Power draw/ battery life less power draw, averages 2–3 watts, resulting in battery boost More power draw, averages 6–7 watts and therefore uses more battery Storage capacity Typically not larger than 1 TB for notebook size drives; 4 max for desktops Typically ~ 500 GB and 2 TB max for notebook size drives; 10 TB max for desktops Cost $0.20 per GB for a 1-TB drive ~. $0.03 per GB for a 4-TB drive

SSD Practical Issues

• SDD performance has a tendency to slow down as the device is

used due to a high level of fragmentation.

• A flash memory becomes unusable after a certain number of

writes.

• As flash cells are stressed, they lose their ability to record and

retain values.

• Most flash devices are capable of estimating their own

remaining lifetimes so systems can anticipate failures.

Outline

 Magnetic Disc  RAID  Solid State Drives  Optical Memory  Magnetic Tape

Optical Storage CD

 Compact Disc

 Originally for audio  650Mbytes giving over 70 minutes audio  Polycarbonate coated with highly reflective coat, usually

aluminium

 Data stored as pits  Read by reflecting laser  Constant packing density  Constant linear velocity (CLV)

CD Drive Speeds

 Audio is single speed

 Constant linear velocity=1.2 ms-1  Track (spiral) = 5.27km  Gives 73.2 minutes

 Other speeds are quoted

as multiples e.g. 24x

 Quoted figure is maximum drive can achieve

CD Operation

CD - Single Layer

CD Format

Mode 0=blank data field Mode 1=2048 byte data+error correction Mode 2=2336 byte data

CD block format

CD - Products

 CD (audio)  CD-ROM (data non erasable)  CD – R (data recordable – just once)  CD-RW (dara recordable – multiple times)

DVD - what’s in a name?

Digital Video Disc

Used to indicate a player for movies

 Only plays video discs

Digital Versatile Disc

Used to indicate a computer drive

 Will read computer discs and play video discs

DVD - technology

 Very high capacity (4.7G per layer)

Bits packed more closely

 Spacing between loops (1.5 μm → 0.74 μm)  Distance between pits (0.834 μm → 0.4 μm)

Up to two layer per side May be two sided

 133 min video

Using MPEG compression (otherwise 4 min)

DVD - Dual Layer

How a DVD works

DVD - Capacity

Name

DVD

Region Codes

Blu-Ray

Essentially the same DVD architecture

* 6 hour of Full HDV - video

Blu ray HD DVD DVD Capacity (Gb) 25/26 (single)* 50/54 (double) 15 (single) 30 (double) 4,7 (single) 8,5 (double) wavelength 405 nm 400 nm 650 nm Transfer rate 54 Mbps 36 Mbps 11 Mbps Scratch resistant yes no no formats

Optical Memory Characteristics

Blu-Ray

Código de Regiões

Outline

 Magnetic Disc  RAID  Solid State Drives  Optical Memory  Magnetic Tape

Magnetic Tape

 Same reading/recording as disc systems  Serial access  Slow  Very cheap (1/3)  High durability (up to 30 years)  Backup and archive  Linear Tape-Open (LTO) Tape Drives

 Developed late 1990s  Open source alternative to proprietary tape systems

LTO Tape Format

 Bands:

 Guard – no data  Data – data  Servo – location info

 Lossless compression  Encryption  Serpentine recording

LTO Ultrium Roadmap

LTO-6 Driver

Tape Libraries

Text Book References

These topics are covered in

Stallings

• chapter 7

External Memory

END