Download Input/Output Systems Input / Output Systems Input / Output Systems I

Input / Output Systems
 CPU and Memory System performance are only part
of the picture, consider the following scenario: Input/Output Systems
 Your system’s a dog. You need to improve its performance.
 On average your system spends about 75% of its time using
the CPU and 25% of its time doing I/O. There are several
options available:
 A new multiprocessor upgrade:
COMP 251
Computer Organization and Architecture
Fall 2009
•  60% improvement in CPU performance for $11,000
 A new I/O system:
•  55% improvement in I/O performance for $4,500
 Which upgrade gives you the most bang for your buck?
Input / Output Systems
I/O System Concepts
 The I/O System: A subsystem of components that
transfer data between the core system and external
devices.  Data Transmission: The ways in which data can be
transferred between devices in a system.
 Serial
 Parallel
 Core System:  CPU / Cache / Main Memory
 Device Type: Indicates the way that a device
physically reads/writes data.
 External Devices (Peripherals, I/O devices):
 Disk Drives
 Graphics Card
 Network Card
 Keyboard / Mouse  Etc…
 Character devices
 Block devices
Serial Communication
Parallel Communication
 Serial: a single bit of data is transmitted at a time.
 Parallel: many bits of data are transmitted
simultaneously.
 Requires 2 wires + control lines
 Each bit is transmitted as a voltage differential
(Vcc
typically 5 volts)
+vcc
wire1
wire2
1
0
0
1
0
1
1
Ground
-vcc
newest
oldest
 Requires 1 wire for each bit of data
 Each bit is transmitted as +Vcc (1) or Ground (0).
 Electrical interference causes signals to degrade with
distance
 High speed parallel communications are typically limited to a few
feet.
 Examples:
 Examples:
 PC Serial Port (RS232/332)
 USB  Fire Wire (IEEE 1394)  Ethernet
Receiver
Sender
•  1 Wire 1 = +Vcc, Wire 2 = -Vcc •  0 Wire 1 = -Vcc, Wire 2 = +Vcc
12 Mbs (1.1)
480 Mbs (2.0)
400 or 800 Mbs
10 or 100 or 1000 Mbs
 ATA / IDE
 SCSI  PC Printer Port
100 MB/s
80 MB/s
1
Character Devices
Block Devices
 Character devices read / write a single byte (or
word) of data at a time.
 Block devices read / write a collection of bytes
(or words) at a time.
 Examples:
 Main Memory
 Keyboard / Mouse
 Network Interface Card
 USB / Flash Drive?
Don’t Try This At Home
 What do you get if you stuff your computer’s
disk drive with herbs?
 A thyme machine
Physical Devices
 Physical devices are often complicated
electromechanical machines.
 Each brand of device has its own peculiarities.
 E.g.  Disk drive
 Monitor: CRT / LCD
 Pointing Device: Mouse / touch pad / track ball
 Example:
 Disk drive
 CD / DVD ROM
 USB / Flash Drive?
I / O System Components
 The main components in an I/O System are:
 System Bus
 BIOS / Device Drivers
 Interface cards (a.k.a. adapters)
 Physical I/O Devices
Interface Card
 An Interface Card provides a uniform programmatic
interface to a physical device:
 All interface cards for a type of device (graphics, disk
drive, etc) support a standard basic set of instructions for
interacting with that type of device.  Cards from specific manufacturers typically also support enhanced
or expanded instructions that are specific to that manufacturer’s
device.
 Instructions are issued by writing into an instruction
register on the interface card.  Other registers on the card get loaded with parameters for the
instruction or are read to obtain the results of an instruction.
2
Basic Interface Card Operations
 Examples:
 Disk Interface:  Read block 22
 Write block 74
 Graphics Interface:  Color the pixel at location (20,30) using the color light blue.
 Place the character ‘A’ at location (10,15)
BIOS: Basic I/O System
 The BIOS contains a set of machine language subroutines for issuing basic instructions to interface
cards.
 Programs can call these sub-routines to carry out basic
input/output operations.
 Hides the complexity of reading and writing device
registers behind the standard procedure calling interface.
 Keyboard Interface:
 Read the last keystroke.
 Mouse:
 Read the last mouse action (left click/right click etc…).
 BIOS also contains routines for editing the basic
configuration of the computer and for bootstrapping
the operating system.
Bootstrapping
 Bootstrapping is the process by which the computer
loads its operating system when it is turned on.
 Typical bootstrap process:
 PC initialized to address of a small program in BIOS Bootstrap
routine
•  Loads first sector of the boot disk into a known memory location.
•  Contains a small program (~512 bytes) – the primary bootstrap
loader.
 Bootstrap routine JUMPs to first instruction of the primary
bootstrap loader.
 Primary bootstrap loader reads a larger more powerful program –
the secondary bootstrap loader - from a specified location on disk Device Drivers
 Device Driver: Used by the operating system,
in place of BIOS routines, to interact with
devices.
 Device drivers can utilize the enhanced or
expanded instruction set provided by the specific
physical device that is present.
 Take advantage of additional features:
 E.g. Graphics cards with 2D or 3D support.
 Use the device more efficiently:
 E.g. Schedule disk read/write operations.
Methods for Performing I/O
 There are four common methods that are used
by BIOS routines and device drivers to
perform I/O operations.
 Each method is designed to decrease the
involvement of the CPU in the I/O operation.  Programmed I/O
 Interrupt Driven I/O
 Direct Memory Access (DMA)
 Channel I/O
Programmed I/O
 In programmed I/O the CPU explicitly controls all
aspects of the the input and output of the data.
 Disk Read Example:  CPU requests the first block of data of the file from the disk
 CPU waits for disk drive to read the block
 CPU transfers block of data to main memory word by word.
 Repeat for each block of data in the file
 Used for:
 Reading and writing main memory.
•  E.g. Copying one array into another array.
 Dedicated devices
•  ATM
•  Alarm System (polling)
3
Interrupt Driven I/O
 In interrupt driven I/O the CPU is free to perform
other work while a device is reading/writing the data.
  Disk Read Example:  CPU requests the first block of the file
 Disk drive reads the first block into a buffer. Meanwhile the CPU is free to do other work.
 Disk generates an interrupt when the block has been read into the buffer.
 CPU stops (interrupts) what it is doing and transfers the data in the buffer to main memory word by word.
 Repeat for each block of the file.
Interrupt Driven I/O
 What tradeoff are we making?
 Speed vs. Cost/Complexity
 When is is most useful?
 When the device is significantly slower than the
CPU (e.g. disk drive / network / printer).
 When the the availability of data is initiated by an
external entity (e.g. keyboard / mouse / network).
Cycle Stealing
 Cycle Stealing: Occurs when both the CPU and the
DMA controller need to access main memory
 Only one device at a time can access main memory.
 The DMA controller “steals” memory cycles from the CPU
•  Priority for main memory access is given to the DMA controller.
•  Prevents buffer overflows and lost data.
•  E.g. data transfer over the network interface.
 Cycle stealing is the reason for sluggish system performance during
periods of high I/O (e.g. printing, downloading).
•  Cache helps
Double Buffering
 The performance of interrupt driven I/O can be
improved by using two buffers.
 Only one device, the Disk or memory can use the buffer at
a time.
 Using two buffers allows both the disk and the CPU to be
busy simultaneously.
Direct Memory Access (DMA)
 With direct memory access a DMA controller handles
the transfer of data from a device buffers to memory.
 Disk Read Example:
 CPU instructs DMA controller to read first block of the file from disk and store it into main memory at a specified address.
 CPU goes off to do other work.
 DMA controller requests first block of file from disk.
 Disk reads the first block into a buffer and notifies the DMA controller.
 The DMA controller transfers the block from the disk buffer into main memory.
 The DMA controller generates
an interrupt for the CPU.
 Repeat for each block of the file.
Channel I/O
 Channel I/O is an extension of DMA in which the
DMA controller becomes a small CPU (an I/O
processor) capable of executing programs.
  The CPU downloads a program to the I/O processor which
carries it out.
 The program may include instructions to transfer multiple blocks
of data from multiple files to different locations in memory.
 In the case of CD Jukeboxes or robotic tape libraries, the program
may also contain instructions for switching CDs and rewinding
tapes.
 Expensive
 Used primarily on high performance main-frames.
4
Interrupt Processing
 When the CPU receives an interrupt the
operating system is invoked to process the
interrupt.
Polled Interrupt Processing
 In polled interrupt processing all devices share a
single interrupt request line.
 Two approaches:
 Polled Interrupt Processing
 Vectored Interrupt Processing
Vectored Interrupt Processing
  In vectored interrupt processing an interrupt controller is used
to identify the device that generated the interrupt.
  The interrupt request line is one of the wires in the control portion of
the system bus.
  A device sets the interrupt request line to 1 (asserts it) when it wants to
interrupt the CPU.
  When the interrupt occurs the OS is invoked.
  The OS polls each device to determine which generated the interrupt.
  OS invokes appropriate sub-routine / device
driver to handle the interrupt.
Persistent Storage Technology
  The Early Days:
  Punch Cards / Paper Tape
  Magnetic Tape
  Hard Disks
 IBM, 1956
 ~5MB, 1 second access time
 $1,000,000   A device asserts its interrupt request line.
  The interrupt controller asserts the INT line to the CPU.
  The CPU acknowledges the interrupt by asserting the ACK line.
  The CPU reads the IRQ number of the device that caused the interrupt.
  The IRQ number is used as an index into a table (interrupt vector).
  Vector contains the address of the OS sub-routine / device driver for
handling the interrupt.
  PC is set to address of the found in the table.   Today:
  HDD:
 SCSI: 1TB, 3Gb/s, 16MB buffer, 4.16ms access, 7.2k RPM, ~$230
 ATA: 1TB, 300MB/s, 32MB buffer, 4.2 ms access, 7.2k RPM, ~$150
 USB: 1TB, 480Mb/s, 7.2k RPM, ~$100
  Optical:  4.7GB (Single-sided, Single Layer DVD)
 17GB (Double-sided, Double Layer DVD) Hard Disk Structure
Hard Disk Structure
  Tracks
  Cylinders
  Sectors
  Interleaving
  Blocks/Clusters
  Disks read/write a block
of data at a time.
  Zoned Bit Recording
5
Hard Disk Capacity
 What is the total capacity of a hard disk with:
 4 platters
 1024 tracks per surface
 64 sectors per track
 512 bytes per sector
Optical Disk Structure
 Optical Disks:
  CDROM, CDR, CDRW, DVD, DVDR, DVDRW, WORM
Hard Disk Access
 Addressing Data:
 C:H:S: Cylinder Head Sector addressing
 Logical: Sectors numbered from 0 to N.
 Access Time: The time for the disk to begin reading
from a specified address.
 Seek Time: Time required for the heads to move to the
requested cylinder.
 Rotational Latency: Time required for the requested sector
to rotate under the read/write head.
 Average values are typically reported.
Optical Disk Data Encoding
 CD:
1 micron = 1000 nm  Pit: a bump on the face of the disk.
 Each pit edge encodes a 1.
 All other regions represent 0’s.
•  On a CD every 0.83 microns is a bit.
 Land: space between the pits.
RAID
 RAID ≡ Redundant Arrays of Inexpensive Disks
 A.K.A. Redundant Arrays of Independent Disks
 The idea of RAID is to use multiple disk drives to increase
the performance and/or reliability of secondary storage
systems.
 Higher throughput
 Disk failure recovery
 Original RAID description had 5 levels:
•  RAID-1 … RAID 5
•  Not all are commercially available.
RAID Concepts
  Data Striping: Spreading data across multiple disks to allow
for concurrent reading of different parts of the data.
  Data Mirroring: Storing copies of the same data on multiple
disks for drive failure recovery.
  Parity: Storing extra bits of data for error detection and
additional fault tolerance.
  Error Correcting Codes (ECC): Storing extra bits of data for
error correction and additional fault tolerance.
 Two additional levels have been added:
•  RAID-0 … RAID 6
6
RAID-1
RAID 0
 RAID-0: Block Interleave Data Striping
 RAID-1: Disk Mirroring
  Redundancy
 Not really RAID - no redundancy
 High Performance:
 Data striping allows for multiple concurrent read and write
operations
 High Bandwidth Applications
•  Audio/Video Production & Editing
•  Image Editing
 Tolerates any single drive failure
 Tolerates certain simultaneous drive failures
  Some concurrent read/write operations
  Large hardware overhead
 Requires 2N disks, N = # of disks needed to store 1 copy of data.
  Fast recovery
  High Availability Applications
 Accounting, Ecommerce, Payroll
RAID-3
RAID-4
 RAID-3: Bit Interleave Data Striping with Parity Check
 RAID-4: Block Interleave Data Striping with Parity Disk
 Just like RAID-3 but stripe blocks instead of bytes.
 Redundancy:
 Tolerates any single disk failure.
•  Continuous operation with failed disk.
 Lower cost than RAID-1 (requires only N+1 disks)
 High bandwidth applications (requiring redundancy)
 Best for block data (E.g. Image/Video)
RAID-5
  RAID-5: Block Interleave Data Striping with Distributed Parity
  Redundancy:
 Can tolerate any single disk failure.
  Same cost as RAID-3
  Better random read / write performance.
  Most widely available RAID implementation
 File/Application Severs
 WWW/E-Mail Severs
 Database Servers
RAID-2
 RAID-2: Bit Interleave Data Striping with Error Correcting Hamming Code
  Redundancy:
 Tolerates the loss of any single disk.
 On the fly error correction.
  Expensive:
 Requires large number of disks.
7
Hamming Codes
 Hamming Code: An error correcting code based on
Hamming Distance.
 Hamming Distance: The number of bit differences
between two binary values:
 Example:
1011
0010
•  Hamming Distance = 2
 Basic Idea: Encode data using only values that are at least a
hamming distance of three apart. Then if any one bit is in
error correct it to the nearest value.
Using Hamming Codes
 Consider the following set of Hamming code
values with Hamming distance 3:
 00000
 01011
 10110
 11101
 Imagine we read the value 10000.
 This is not a valid value in our Hamming code.
 How do we correct it?
Other RAID Systems
 RAID-6: Block Interleaved Data Striping with
Multiple Independent Distributed Parity Schemes
 Tolerates multiple concurrent disk failures.
 Requires N+2 disks.
 Nested RAID Levels
 Combinations of multiple RAID Levels
 RAID-01: Mirrored array of striped disks.
 RAID-10: Striped array of mirrored disks.
 Yet others: •  03, 30, 53, 05, 50, 15, 51
Nested RAID
RAID-10
Striped array
of Mirrored
Disks
RAID 0
/-----------------------------------\
|
|
|
RAID 1
RAID 1
RAID 1
/--------\
/--------\
/--------\
|
|
|
|
|
|
120 GB
120 GB
120 GB
120 GB
120 GB
120 GB
A1
A1
A2
A2
A3
A3
A4
A4
A5
A5
A6
A6
B1
B1
B2
B2
B3
B3
B4
B4
B5
B5
B6
B6
RAID 1
/--------------------------\
|
|
RAID 0
RAID 0
/-----------------\
/-----------------\
|
|
|
|
|
|
120 GB
120 GB
120 GB
120 GB
120 GB
120 GB
A1
A2
A3
A1
A2
A3
A4
A5
A6
A4
A5
A6
B1
B2
B3
B1
B2
B3
B4
B5
B6
B4
B5
B6
RAID-01
Mirrored array
of Striped Disks
8