Input/Output

Download Report

Transcript Input/Output

Input/Output
The computer’s response time is no
match for ours.
Presentation
dr. Patrick De Causmaecker
url: http://www2.kahosl.be/~patdc
E-mail [email protected]
Books & reference
• Modern Operating Systems, A. Tanenbaum,
Prentice Hall inc. 1992, chapter 5 (main text
and exam material)
• Operating Systems: Internals and Design
Principles, third edition, Prentice Hall inc.
1998, chapter 11. (more detailed text for
reference and further reading)
• http://WilliamStallings.com/OS3e.html
The O.S. and the I/O hardware
• Management of I/O devices is a main task for the
operating system.
–
–
–
–
Handle message passing to and from the system
Handle interrupts and errors
Offer a simple interface
Offer device independence (to some degree)
• Viewpoints.
– Users
– Programmers
– Technicians
Overview
•
•
•
•
•
•
I/O Devices
Principles of I/O Hardware
Design Objectives
Disks
Clocks
Terminals
I/O Devices
• (Classification Stallings)
• Man readable
– Printers, graphical terminals, screen, keyboard,
mouse, pointing devices,…
• Machine readable
– Disks, tapes, sensors, controllers, actuators
• Communication
– Modems
I/O Devices
• Differences
– Data transfer rates
• Range from 0,01 kbytes/sec (keyboard) till 30.000
kbytes/second (graphical screen)
• Disks typically have 500 kbytes/sec (optical) till
2000 kbytes/sec (magnetic)
– Application
• Disk for files or for virtual memory
• Influences disk scheduling and caching strategy
– Complexity
• OS is isolated from most of the complexity by an
I/O module (controller)
I/O Devices
– Unit of transfer
• Blocks of characters versus streams of characters
– Representation of data
• Coding conventions, parity,…
– Error states
• Kind of errors, reporting modes,…
• Need a uniform approach to I/O as seen
from the user and from the operating system
point of view
Organisation
• The I/O function can be approached in three
ways:
• Programmed I/O: continuous attention of the
processor is required
• Interrupt driven I/O: processor launches I/O and can
continue until interrupted
• Direct memory access: the dma module governs the
exchange of data between the I/O unit and the main
memory
Programmed I/O
Memory
Databus
CPU
I/O
I/O
Programmed I/O (Polling)
&
Int A?
j
Service
routine A
&
Address
Decoder
Int C?
&
Int Z?
j
Service
routine C
j
Service
routine Z
Interrupt
Memory
Databus
CPU
Interrupt
I/O
I/O
Interrupt controlled I/O
Data
CPU
Interface
Devices
Int acknowledge
2
Interrupt
Program
1
Program
Interruptroutine
CPU
I/O
I/O
Direct Memory Access
Memory
Databus
CPU
hold
I/O
DMA
I/O
Direct Memory Access
databus
Hold
CPU
Ack
DMA
Ram
I/O
addressbus
R/W
Evolution in I/O organisation
1. Processor controls device directly
2. A controller or I/O module is used separating the
processor from the details of the I/O device
3. As in 2 but using interrupts, the burden of
supervising the device continuously disappears
4. I/O module gains access to main memory
through DMA moving data in and out memory
without processor attention
5. I/O module becomes a processor - I/O specific
instruction set – to be programmed by the
processor in main memory
6. I/O module becomes a computer – a “channel” -
Direct Memory Access
• I/O module shares the bus with the processor
• It can use the bus only when the CPU does not
need it
• Or it can steal cycles from the CPU by forcing it
to free the bus
• Procedure:
– Processor sends message to DMA (R/W, where, how
much)
– Processor continues while I/O proceeds
– At the end, I/O module sends interrupt signal to the
processor
Differs from an interrupt
• DMA can interrupt the processor in the
middle of an instruction (before fetch, after
decoding, after execution)
• It does not interfere with the running
program
• The CPU is idled (waiting for access to the
bus)
• The system is just becoming slower
Three implementations
• DMA shares bus with I/O units
cpu
dma
I/O
I/O
memory
• Performance and simplicity in design I/O module
Three implementations
• DMA shares bus only with other DMA, CPU,
memory
cpu
dma
I/O
I/O
memory
• Performance and simplicity in design I/O module
Three implementations
• DMA has a separate I/O bus
cpu
dma
memory
I/O
I/O
• Performance and simplicity in design I/O module
Overview
•
•
•
•
•
•
I/O Devices
Principles of I/O Hardware
Design Objectives
Disks
Clocks
Terminals
Principles of I/O Hardware
• Block Devices
–
–
–
–
–
Store information in fixed sized blocks
Size 128-1024 bytes.
Blocks are the addressing units for reading and writing
Example: disks
More problematic example: tapes
•
•
•
•
Not randomly accessible
Rewind takes far more time than a seek on a disk
Writing in the middle of a tape is not always possible
Discussion is not really a practical one
Principles of I/O Hardware (cont)
• Character Devices
– Terminal, printer, network interface, mice,
rats...
– Token stream in and out
– No addressing possibility, no seek.
Principles of I/O Hardware (cont)
• Blocks and character devices are still a basic
abstraction at the O.S. level (file systems)
• Not everything fits:
– Clocks are not addressable, do not produce token
streams, produce interrupts instead
– memory mapped devices are character addressable
• Physical details and peculiarities are left to the
Drivers
Hardware Principles: controllers
• I/O units have electronic and mechanical
components separated in modular design.
• The electronic component is the “device
controller” or “adapter”
–
–
–
–
–
CPU
Print to be plugged into the computer
Designed independently for many devices
Standardised interfaces (ANSI,ISO,IEEE,…)
Is the target for the O.S.
Bus <-> I/O channels
Mem
Disk
Prtr
bus
Principles: controllers (cont)
• Controller-device interface
– Very low-level
– Example: disk
•
•
•
•
Preamble with cylinder, sector no,…
4096 bits from this sector
Error correction code
Controller produces a block of bytes and performs
error correction if necessary, copies block into
memory
Principles: controllers (cont)
– Example: video card
• Reads bytes from own memory and generates signals to steer
the CRT
• Programming the electron rays is clearly unfeasible for normal
programmers
• Controller must offer an abstract interface
• Controller-CPU interface
– Based on controller registers
– These may be part of the normal address space of the
computer (memory mapped I/O) (68x0)
Principles: controllers(cont)
• Specific address space for I/O (Intel ports IN and
OUT instructions)
• In both cases the decoding logic on the card
recognizes the addresses
• They are normally served by system calls
(software interrupts) supporting ‘abstract’
commands such as READ, WRITE, SEEK,
FORMAT, RECALIBRATE
• CPU passes commands to the controller and
continues. It will be interrupted when the
transaction has finished
Hardware Principles: DMA
• Programmed I/O asks for too much attention of the
CPU if the device is fast
• Fast devices use Direct Memory Access
• Without DMA:
–
–
–
–
–
CPU
Controller reads block into an internal buffer
Checks for errors, corrects eventually
Causes interrupt
CPU copies block into memory through card registers
Controller starts reading again
Mem
Disk buffer
Controller
start
teller
Hardware Principles: DMA
(cont)
• With DMA:
• Controller keeps a memory address (called the DMA
address) and a counter indicating the number of
bytes to be copied in addition to the disk address of
the block(s) requested.
• Once block is in the controller buffer, the controller
copies the bytes one by one to the DMA address,
decrementing the counter and incrementing the
DMA address
• Only when the counter becomes zero an interrupt is
generated
Hardware Principles: DMA
(cont)
• Why a buffer?
– Hardware produces bits at a steady pace
– Controller has to share the system bus with the
CPU and other devices, unpredictable
– Copying is not time critical
– Controllers with small buffers may go directly
to the memory, (overrun risk)
– Buffer simplifies the design of the controller
• Controller processing time leads to
interleaving
Hardware Principles: DMA
(cont)
• Interleaving on disks
– 3600 rpm, revolution time 16 ms
– This time can be used for error correction and
(start up of) DMA transfer
– Room between successive sectors can be
adjusted by interleaving
7 0
6
1
54 32
7 0
3
4
62 51
5 0
2
3
74 16
Overview
•
•
•
•
•
•
I/O Devices
Principles of I/O Hardware
Design Objecives
Disks
Clocks
Terminals
Design objectives
• Efficiency
– I/O is very slow in comparison with the CPU
– Multiprogramming is partly invented to
circumvent this discrepancy
– Even with a high level of multiprogramming,
the I/O cannot cope with the CPU
– Swapping is not I/O but it sometimes interferes
with it
– Disk efficiency is important
Design objectives
• Generic
– For simplicity and to avoid errors
– Uniform treatment of all devices
– Due to the diversity of the devices, only layered
system can offer this uniformity
– Layers depend on lower layers for specific
services, implement general set of services
themselves
– Number of layers can be different from one
kind of devices to another
Design objectives: three models
User
Processes
User
Processes
User
Processes
Directory
Management
Logical
I/O
Comm.
Architecture
File
System
Physical
Organization
Device
I/O
Scheduling
& Control
Hardware
Local peripheral device
Device
I/O
Scheduling
& Control
Hardware
Communications port
Device
I/O
Scheduling
& Control
Hardware
File System
Design objectives
• Simple case: logical device, offering only a
stream of bytes or records
– Logical I/O: has nothing to do with details of
the device, offers logical I/O functions
– Device I/O: translates commands from the layer
above to sequences of I/O instructions.
Buffering may be part of this layer
– Scheduling and controlling: this level
determines the order in which the requests can
be served. This layer interacts directly with the
I/O device
Design objectives
• Communication: network interface, offering
connection to a network
– Communications layer: replaces the logical
layer. It may be layered itself according to e.g.
the OSI seven layer model
– Other layers are identical with the logical case
Design objectives
• File management
– Only the first layer is different
– It is normally layered itself in
• Directory management
– Symbolic filenames are converted to file references
– Supports moving, deleting, creating… files
• File management
– Logical file structure for open, close, read, write,…
• Physical registry
– File in sectors, tracks,…
Efficiency: I/O Buffering
• I/O module may free the process of
continuously checking by working
autonomously
• If the process is swapped out, the dma
cannot work, so swapping of the complete
process becomes problematic
• One process can even cause a deadlock
• -> buffering (single buffer, double buffer,
circular, buffer)
I/O Buffering: single buffer
• I/O module copies to a buffer, not directly to the
memory.
• Once I/O has succeeded, buffer is copied to the
user memory
• I/O module may anticipate and start reading the
next block immediately
• Complicates OS
– Has to manage the buffers
– Has to take care while swapping to the same disk
(swapping out a process has to wait until I/O has
finished)
• Buffers can contain blocks, lines or bytes
I/O Buffering: double buffer
• Without buffers, the time per block is
– T = Computation + Transfer
• The time for reading a block with a single
buffer:
– T = max{Computation, Transfer} + Move
• The Move time to copy the buffer into main
memory can be recovered if one uses a
second buffer, keeping the I/O channel busy
all the time
– T = max{Computation+Move, Transfer}
I/O Buffering: circular buffer
• If the computation time in I/O burst phases
is very small, the two buffers will not be
able to cope (max{C,T} >> C)
• Multiple buffers in a ring can solve this
problem by doing I/O also in the
computational phases
• Buffering softens the peaks in the I/O. If the
demand for I/O is on average above the I/O
capacity, the system will run out and the
process will have to wait.
Generic : Software principles
• Layered architecture allows for
– Hiding of the peculiarities of the hardware
– Presentation of clean interfaces for the users
– Construction of uniform interfaces
• After stating the objectives, we will discuss
the layers from the interrupt level up to the
user environment
From “Mobile Commerce, a new frontier”,
V. Upkar, R.J. Vetter, R.Kalakota in Computer, October 2000
Software Objectives By Example
• Device independence: programs must be able to
access files through a device independent interface
– sort < input > output
– Naming conventions cannot depend on the device
– Unix devices can be reached through the directory
interface (mounting)
• Error handling: errors should be taken care of in
the lowest layer possible (controller-driver-osuser)
– Errors are often volatile, caused by a temporary defect
– They can be handled without alarm to the user
Software Objectives By Example
• Device independent interface
(UNIX/LINUX)
sort < input > output
(Naming conventions cannot depend on the
device)
– /dev/lp
– /dev/d001
(Unix devices can be reached through the
directory interface (mounting))
Software: how to cope with
interrupts
• “Hard” part of the software,
– Should be kept in the basements
• Essentially pseudo parallel
– Should be synchronized with the user process
• Semaphores and monitors can be used to
model the interaction with the I/O process
– E.g. socket threads
• Messaging.
Example: City Dataservice
Citizen
Service Shop
Task Based Request
Handler
LAN
interface
Dispatcher
Tasks
Governmental
Database Server
WAN
interface
Software: device drivers
• One type or at most one class of closely related
devices.
• Contains all device dependent code.
• E.g. one terminal driver serving all terminals is not
feasible: complexity of terminals can differ too
much.
• Is the only part in the OS that knows the
controllers, their register structure, sectors, tracks,
cylinders, heads, interleaving, motors, ...
• Device driver translates abstract, device
independent software requests in device dependent
signals.
Software: device drivers (cont)
• Example: read a block from disk
– If driver is idle when request arrives
• Start immediately
– Else
• Move request to queue
• Execute request:
– Do a translation to the disk specifics
• Find position of the block
• Check whether disk is rotating
• Check arm position
– Pass tasks to device one by one
• Some controllers can accept a list of tasks, others need
assistance of the driver after each subtask
Software: device drivers (cont)
• Once the task(s) have been passed:
– Two possibilities
• Driver has to wait until the controller has performed the
task(s), in which case the driver has to block until interrupted
– Example: disk I/O, scanner
• Or the controller can finish the job without delay
– Example: move the graphics over the screen
• Driver has to check on possible errors before
passing eventual results and/or status to the
software
• Once all software tasks have been handled the
driver will block
Software: Device Independence
• Large part of the I/O software is independent of
the device. Some of these parts are executed in the
driver for efficiency (cfr LINUX & Apache)
• Independent parts normally include
–
–
–
–
–
–
–
–
Uniform interfaces for the device drivers
Naming of the devices
Protection and privacy for devices
Block size independent of the device
Buffering
Memory assignment on block devices
Assignment of dedicated devices
Error handling
Software: Device Independence
Uniform interfaces
• Software should not be aware of detail
differences between the devices
• Users must be able to handle a disk as a
disk and a scanner as a scanner, not work
with a “Brother Master Scan” or a “Sister
LaserJet” but with an A4 scanner and a
black and white printer.
Software: Device Independence
Naming of the devices
• How do files and I/O devices get a name
• Symbolic device name is mapped to a driver
• E.g. UNIX/LINUX:
– Device name: /dev/lp determines I-node for
special file
– I-node contains
• Major device number to find the driver
• Minor device number to be passed as a parameter to
the driver
Software: Device Independence
Protection and privacy
• Users access to devices must be regulated
– No limitations
• MS DOS and early versions of Windows
– No access
• Mainframe and certain minicomputer O.S.
– Way between
• UNIX: rwx bits
Software: Device Independence
Device independent block size
• Communication block sizes may differ
– Software provides one size, handling a set of
sectors as one block
– Devices become abstractions
– Block devices can be represented as character
devices
Software: Device Independence
Buffering
• Block devices:
– Hardware delivers one block at a time
– Software reads in varying sizes
– After reading one block, the O.S. keeps the whole block
in a buffer
• Character devices
– May be slower than the processes
– Delivery of characters may be an independent process
(keyboard)
Software: Device Independence
Assignment of dedicated devices
• Some devices cannot be used by two
processes at the same time
– OPEN system call reports an error in case the
device is busy
– CLOSE frees the device
– Spooling processes may play an in between role
Software: Device Independence
Error handling
• Primary handling is by the drivers
– It can often react and repair the error before the
software layer above notices
– Only when, after several tries, repairing fails, the
higher layers are informed
• Error handling at higher levels depends on
the environment in which the software is
functioning
I/O software in user space
• I/O software is not only in the OS, libraries are
routinely linked with user programs, helper
programs run outside the kernel.
• count = write(fd, buffer, nbytes)
• Library routines typically transfer their parameters
to the system calls, or do more work:
• printf(“%d : %s, %ld\n”, count, person, yearly_income);
• cout << count << “ : “ << person << “, “ << y_i << endl;
• cin >> y_i;
• Spooling systems: a daemon reads jobs from a
spooling-directory and sends it to the printer
(protected device, only spooler has access)
I/O software in user space (cont)
• Spooling may be used in networks, mailing
systems,…, spoolers run outside the O.S.
• Overview:
user processes
Device independent
software
Device drivers
Interrupt handlers
hardware
Produce I/O call, format, spool
Naming, protection, blocking,
buffering, allocating
Control device registers, status
Wake up driver after I/O task
Perform I/O
Sketching Interfaces: Toward More Human Interface
Design
James A. Landay—University of California, Berkeley
Brad A. Myers—Carnegie Mellon University
(Computer, march 2001)
Overview
•
•
•
•
•
•
I/O Devices
Principles of I/O Hardware
Design Objecives
Disks
Clocks
Terminals
Disks
• Rationale: why an additional memory
carrier
– Larger available capacity
– Lower price per bit
– Permanent
• Hardware and algorithms
Hardware
• Disks are organized in cylinders, tracks and
sectors.
• Sectors
– between 8 and 32
– equal number of bytes (1K, 0.5K,0.25K) in the middle
as well as on the border
• Controller can run a number of seeks on different
disks at the same time
• Controllers only read/write one disk at the time
• This influences the disk driver
Algorithms for the disk arm
• Access time =
seek time : time needed to move the arm to the
cylinder (dominant)
+ rotational delay : time before the sector appears
under the head
+ transfer time : time to transfer the data
• Dominance of seek time leaves room for
optimisation
Example : seek time
• Time to get the arm over the track, difficult
to determine
Ts = m*n + s
– Ts = estimated seek time
– n = number of tracks crossed
– s = start time
• Cheap: m=0,3, s=20ms
• Better: m=0,1, s=3ms
Example: rotational latency and
transfer time
• 3600 rpm, 16,7 ms/rotation
• Average rotational latency is 8,3 ms
Tt=b/rN
–
–
–
–
Tt = transfer time
b = number of bytes to be transferred
N = number of bytes per track
r = rotation speed in rotations per second
Example: case
• Read a file of size 256 sectors with
– Ts = 20 ms
– Tt = 1 MB/second
– 512 bytes/sector, 32 sectors/track
• Suppose the file is stored as compact as possible:
all sectors on 8 consecutive tracks of 32 sectors
each (sequential storage)
• The first track takes 20 + 8,3 + 16,7 = 45 ms
• The remaining tracks do not need seek time
• Per track we need 8,3 + 16,7 = 25 ms
• Total time 45 + 7*25 = 230 ms = 0,22 s
Example: case
• In case the access is not sequential but at
random for the sectors, we get:
• Time per sector = 20 + 8,3 + 0,5 = 28,8 ms
• Total time 256 sectors = 256*28,8 = 7,37 s
• It is important to obtain an optimal
sequence for the reading of the sectors
Optimisation
• Heavy loaded disk allows for a strategy to
minimize the arm movement
• Situation is dynamical: disk driver keeps a table of
requested sectors per cylinder
• E.g.: while a request for disk 11 is being handled,
requests for 1, 36, 16, 34, 9 and 12 arrive.
• Which one is to be handled after the current
request?
Optimisation (cont)
• In the order of arrival (FCFS) the total length is:
– |11-1|+|1-36|+|36-16|+|16-34|+|34-9|+|9-12|=111
• In the order “shortest seek time first, SSTF” (cfr
shortest job first) we gain 50%:
– |11-12|+|12-9|+|9-16|+|16-1|+|1-34|+|34-36|=61
• Problem: starvation, arm stays in the middle of the
disk in case of heavy load, edge cylinders are
poorly served, the strategy is unfair (cfr elevators)
• “Lift algorithm, SCAN” : keep moving in the
same direction until no requests ahead (start Up):
– |11-12|+|12-16|+|16-34|+|34-36|+|36-9|+|9-1|=60
• Upper limit: 2*number of cylinders
Optimisation (cont)
• Smaller variance is reached by moving the arm in
one direction, always returning to the lowest
number at the end of the road (CSCAN):
– |11-12|+|12-16|+|16-34|+|34-36|+|36-1|+|1-9|=68
• Seeks are cylinder by cylinder:
– a number of tracks
• It may happen that the arms sticks to a cylinder.
– N-step-SCAN: segment the queue into segments of N
requests which are handled using SCAN
– FSCAN: use two queues, while one is being handled
with SCAN, the other one is being refilled.
Optimisation (cont)
• Sector number under the head can be read
– > optimal order of requests within one cylinder
• More than one drive: look for a well positioned
drive after each transfer
• Rotation speed versus seek time
• RAID (Redundant Array of Inexpensive Disks)
– E.g. 38 bits form a word of 32 bits with Hamming code
– Each bit is stored on a separate disk
Error handling examples
• Possible errors:
– Programming errors (non existing sector)
• User program needs debugging
– Volatile checksum error (dust particle)
• Controller tries again
– Permanent checksum error (bad block)
• Block will be marked “bad” and replaced by a spare block (this
my interfere with the optimisation algorithm)
– Seek error (arm moves to the wrong sector)
• Mechanical problem, perform RECALIBRATE or ask for
maintenance
– Controller error
• Controller is a parallel system, can get confused, driver may
perform a reset
Caches
• The driver may cache tracks:
– Reading one track does not take a long time,
arm needs not to be moved and the driver has to
wait for the sector anyhow
– Disadvantage: driver has to copy the data using
the CPU, while the controller may use DMA
• The controller may build an internal cache:
– Is transparent for the driver
– Data transfer uses DMA
– In this case the driver should not do any
caching
Overview
•
•
•
•
•
•
I/O Devices
Principles of I/O Hardware
Design Objecives
Disks
Clocks
Terminals
Clock Hardware
– 50 Hz clocks (1 interrupt (clock tic) per voltage cycle)
• Simple, cheap, not very accurate, not very functional
– High precision clocks (5-100 MHz, or higher),
•
•
•
•
•
•
Contain a quarts oscillator
Steers a counter counting down
Generates an interrupt when counter reaches 0
Counter is eventually reloaded from a programmable register
One chip normally implements multiple clocks
One shot mode: clock counts down from register value once
and waits for software to start it again
• Block wave mode: counter is automatically reloaded (generates
clock tics)
• Ranges: e.g. 1000 MHz clock with a 16 bits register can fix
time intervals between 1 nanosecond and 65,535
microseconds.
Clock Software
• Software is responsible for the semantics
behind the clock tics:
– Time of the day
• 1/1/1970
• Is an easy task, just calculate the exact time between
two tics and adjust the clock on each interrupt
• Size of the time register may cause a problem:
– 32 bits register overflows after 2 years storing 60 Hz
tics
– 64 bits is more expensive, but lasts forever
– Store seconds in stead of tics (232 seconds is 136 years)
– Use another reference in stead of 1/1/1970 (start time)
Clock Software
– Administration of process time slices
• Each running process has “time left” counter
• This counter is decremented at each interrupt
– Administration of CPU usage
•
•
•
•
Counter starts when process starts
Counter is part of the “Process environment”
Is stopped while handling an interrupt
Field in the process table can be used directly
(through pointer to running process)
• Interrupts cause problems
Clock Software
• SLEEP system call (UNIX)
– E.g. late ack of package sent, sleeping e-student
• Clock driver has a limited number of hardware
clocks
– Implements virtual clocks
– Uses a table with all times for the hanging timers and
one variable with the next signal time
– In case of a heavy clock-usage, the signal times may be
kept in a well ordered linked list (e.g. 4203,4307,4213)
Clock header
3 s1
4 s2
6 s3
Current time
3
Next signal
4200
Clock Software
• Watchdog timers
– Floppy disk drive
• Start motor
• Wait for 500 milliseconds
• -> better to wait for 3 seconds after I/O operation,
just in case a new request arrives
– Watchdog timers start user specified routine
after the time has elapsed within the code of the
caller
• Profiling
– For program performance analysis
– Information where the CPU time is spent on
Overview
•
•
•
•
•
•
I/O Devices
Principles of I/O Hardware
Design Objecives
Disks
Clocks
Terminals
Terminals, hardware
• Serial RS232 terminals
– (~history) hardcopy, glass tty, intelligent
– Industrial applications
• Memory-mapped interfaces
– (~history) character oriented
– Bitmapped
• Network computers
• …
Terminals, hardware (cont)
• Serial interfaces: 25 pins connector (RS232
or V24), needed are 3 pins
• From 50 up to 19200 bits/sec
• Interface: UART (Universal Asynchronous
Receiver Transmitter) on RS232 interface
cards
• Driver writes characters to interface card
which transforms them into a bit sequence
• Slow, interrupt will wake up the driver
Terminals, hardware (cont)
• Interface cards may have a CPU on board,
and be able to serve more terminals
• Intelligent terminals may be able to perform
complicated operations on the screen (e.g.
X-terminals) while still connected through a
prehistoric device as the RS UART
Terminals, hardware (cont)
• Memory mapped terminals
– Integrated into the computer
– Communicate through a special memory, the
video RAM, embedded into the address space
of the CPU
CPU
MEM
Video Ram
Controller
Monitor
bus
Terminals, hardware (cont)
• Monitor talks pixels, video controller modulates
the electron ray or steers the lc’s
• Speed is obtained through the memory and the
electronics
• The ram can be in characters or in bits
– Characters will typically be read 14 times, for each line
on the screen in which part of the character is displayed
– Bits will be stored per pixel
• Input through the keyboard
– Normally passes only a code for the key touched
Input software
• Keyboard driver caches keyboard input
– Raw mode:
• Driver passes characters unchanged to software
• Buffering is limited to speed differences
• Application receives characters immediately
– Cooked mode
• Driver buffers one line until it is finished
• Driver handles corrections made by the user while typing a line
– Often applications have the choice
• Nowadays, window driven applications use raw
mode at the lowest level and perform buffering at
the window level
Input software (cont)
• Keyboard driver catches characters after
– Interrupts: handles character during interrupt
– Messages: message may contain the character or refer
to a small buffer (problem with real time level of the
driver, safer if the messaging is not fail proof)
• Keyboard driver transforms the key number into a
character according to a table
• Keyboard buffering: buffer pool (buffers of equal
size, e.g. 16 characters) or a separate buffer for
each terminal (typically 200 characters)
• Echoing is (was) done by the OS, or the shell.
May be confusing for the user.
Input software (cont)
• Handling of tabs, backspaces,… were typical
problems with terminals
• One problem survived: end of line
– Logically (from the typist’s viewpoint) one needs a CR
to bring the cursor back to the beginning of the line and
a LF to go to the next one
– These two characters are hidden behind the ENTER
key
– The OS can decide how to represent end of line
• UNIX: Line feed only
• DOS: Carriage return and line feed
– LF is ASCII 10, CR is ASCII 13, ^M
• -> problems with file transfer
Input software (cont)
• Other problems with terminals:
–
–
–
–
–
–
–
–
Timing of CR and LF
Padding
Definition of the erase character
Kill character: @ (\@ and \\ conventions)
Stop output (^S) and start output (^Q)
DEL, ^C, stop process (sent SIGINT)
^\ force core dump (SIGQUIT)
^D, ^Z end of file
• Implementation was (is) not straightforward
• UNIX telinfo database contains descriptions of
thousands of terminals (famous example vt100)
Output software
• Serial (RS-232) and memory mapped approaches
differ
• Serial terminals have an output buffer to which
characters are sent until it is full or until a line
ends. Once full, the real output is initiated and the
driver sleeps until interrupted.
• Memory mapped terminals can be accessed
through normal memory addressing procedures.
Some characters receive a special treatment. The
driver is doing more screen manipulation.
• Special functions such as scrolling and animation
may be done through special registers (e.g.
register with the position of the top line)
Example: UNIX/LINUX
Buffer Cache
Hash Table
Hash Pointer
Free List Pointer
Free List
Buffers
Buffered/Unbuffered
• Buffered
– Original UNIX design
– Fast access to disks and other block devices
• Unbuffered
– DMA
– Interferes with system design
– Real time?!
Example : Windows NT
I/O Manager
• Cache Manager
– Whole I/O subsystem (disks, network … )
– Dynamic cache size
• Lazy write
• Lazy commit
• Hardware device drivers
– Uniform naming for source code portability
Asynchronous/Synchronous I/O
• Both are supported in Windows NT
• Asynchronous messaging happens through
– Device kernel object (indicator, one I/O active)
– Event kernel object (threads wait for event)
– Alertable I/O (thread has a queue of ready I/O
operations)
– I/O completion ports (Pool of threads on a
server)
Windows software RAID
• Winows supports RAIDS
– In hardware at the controller level
– In software: non contiguous diskspace is
allocated and combined in logical partitions by
the fault tolerant software disk driver
(FTDISK) It supports RAID 1 and RAID 5
Conclusion
• I/O is complicated because
– of hardware peculiarities (non functional design
decisions)
– it is a real time interface with a non real time
operating system
– it must allow for a broad range of I/O devices
•
•
•
•
•
classical terminals, color screens, X-terminals
hand held devices, mobile phones. PDA’s
printers-scanners, fax
voice interfaces, Braille systems, sensory devices
face recognition systems with camera’s, fingerprint
sensors,…