1. IT253: Computer Organization
Tonga Institute of Higher Education
IT253: Computer Organization
Lecture 13:
Input/Output

2. Input/Output Input and output allows people to work with computers.
It includes many parts: keyboard, mouse, printer, hard drive, network card…

3. I/O Examples

4. Storage Devices The hard disk is an I/O device.
It is long-term storage that is a slow, but still a valuable part of the memory hierarchy

5. Organization of Magnetic Disk
Most of the time, a hard disk will have:
tracks per platter
32 – 128 sectors per track (a sector is the smallest amount of data that can be read or written)
Usually all tracks have same number of sectors, but sometimes they are different because it might be faster to get to one spot on the hard disk than another

6. Characteristics of Magnetic Disks
Cylinder – the word for all the tracks that can be read by the read/write head in the spot that the read/write head is in
Reading and Writing data takes 3 steps
Seek time – move the head to the correct place
Rotational latency – wait for the platter to spin so that the right sector is underneath the head
Transfer time – time it takes to transfer a sector from the head

7. Characteristics of Magnetic Disks
Rotational Latency
Most disks rotate at 3,600 RPM to 10K RPM (rotations per minute). About 3.5 ms to 16 ms for one rotation
Transfer time depends on:
Transfer size (usually 1 KB sectors)
Rotation speed
Recording density (how many bits you can fit in one square inch)
Diameter of the platter
Usually you can get 2 – 12 MB of data transferred a second

8. Disk Access
Access time = queue + seek + rotational + transfer + overhead
Queue – the request has to wait for other things to finish first
Seek time – move arm of head over track
Rotational latency – wait for platter to spin to right place
Transfer time – the number of bits you get a second

9. Disk Access Example Parameters: Transfer size: 8MB
Advertised average seek is 12 ms
Disk spins at 7200 RPM
Transfer rate is 4 MB
Average rotational delay: 4 ms
Controller overhead: 2 ms
What is average disk access time for a sector?
Avg seek + avg rotate delay + transfer time + overhead
12 ms MB/4 MB/s + 2 ms
= 20 ms
Advertised rate was 12 ms, but really 20 ms. The real rate is usually about only ¼ as fast as the advertised rate.

10. Alternative Storage
DRAM as hard disk – you could use DRAM as your hard disk, but you would need the power to be on all the time, and it would be expensive, but very fast
CD-ROM – read only
CD-RW – read and write
Magnetic tape –
Sequential access, very slow, very big in size. Tera or peta bytes of data

11. Buses: Connecting I/O to Processor
A bus is a communication link between devices
It is usually a set of wires that will connect to many things all at once (between different inputs and outputs and the processor and memory)

12. Goods things about Buses
Why do we need buses?
Allows us to add new devices easily
Allows us to move parts from one computer to another if they use same bus type
Lower cost because we only use one set of wires to connect all the pieces

13. Bad things about buses
Buses can also introduce a few problems for computers
They create a communication bottleneck.
Things may not be able to send and get data as fast as possible because it has to wait for other things to use the bus
Maximum speed of a bus is determined by:
Length of bus
How many devices are on the bus
How many devices it has to work with (different devices will have different speeds for sending and getting data

14. Bus Organization
A bus will have two wires, one for control of the bus and one for sending data
Control Lines:
Carries requests and acknowledgements for who wants to use the bus
Tells what kind of data is being sent on the bus
Data Lines:
Carries data between source and destination

15. Master/Slave Bus Organization
To use the bus, two things must happen:
Sending the address
Receiving or sending data
A Master device will start the transaction on the bus by sending the address
A Slave device will respond to this address by either
Sending data to the master
Receiving data from master

16. Outputting Data There are three steps for outputting data
A request on the bus to read memory
Reading the memory
Sending the memory to the output device

17. Input Operation
Input is when the processor receives data from an input device
When we want to do an input operation we need to take two steps.
A request to write memory on the bus
Sending data from input device to memory

18. Types of buses
There are different types of buses that a computer can use to transfer data between different objects
Processor-Memory bus: short and high speed. Connected directly from memory to processor
External I/O bus: longer and slow. Must work with many devices. Connects to processor memory bus
Backplane bus: connects all the pieces together, including memory and processor and all I/O devices. Only one bus needed
Bit-Serial Bus: high speed buses that connect directly from one object to another object. (like USB or FireWire)

19. A Backplane Bus Computer
Backplane bus: a single bus is used to connect all components
Simple and low cost
But it is slow and can cause bottlenecks where data waits for a long time before it gets where it needs to go.
Not used today

20. Two-Bus Computer
A two-bus computer will use an I/O bus for the devices and a processor-memory bus
I/O buses connect to the processor-memory bus through “bus adapters”
The bus adapters allow the different buses to work together without problems

21. Synchronous and Asynchronous Buses
Synchronous bus: this means that the operation of the bus is determined by a clock cycle
It is pretty easy to implement and also runs fast
But every device on the bus must use same clock cycle and if the bus is too long there will be “clock skew”
Clock skew is a problem where a device at the end of the bus may think it’s an older clock cycle, but a new cycle has already started
Asynchronous bus:
No cycle and can work with many devices
It can be any size without worrying about clock skew
Not as fast.

22. Using the bus
One of the most important issues is how will devices get access to control the bus
The master-slave system avoids many of the problems. The bus master controls all actions and tells devices when they will read data and when they will write data to the bus
The simplest way is to make the processor the bus master. That means every time there is an I/O transaction the processor has to be told about it and then tell the bus to do that. This means the processor is slowed down by bus transactions

23. Bus Masters
It is not the best solution to use the processor as the bus master.
Instead, we might have a few pieces of hardware act as the bus masters.
That means there needs to be a way to determine who will use the bus if there are many masters
This is called bus arbitration and it uses two ideas
Priority – if something is really important, then it should be allowed to use the bus right away
Fairness – make sure all the devices are able to use the bus in a reasonable amount of time

24. Bus Arbitration There are generally four bus arbitration ideas used
Distributed arbitration by self-selection - each device on the bus will put a request on the bus to use it when it needs it. It then might have to wait its turn
Distributed arbitration by collision detection – this is what Ethernet uses. It will just send the data on the bus and if it crashes into other data and is lost, then it will send again
Daisy Chain arbitration – there is one device which is a bus arbitrator and it has devices connected one after each other. There is only one wire to connect devices to bus master
Centralized-Parallel arbitration – all devices are connected to a bus master with their own separate lines. Then the bus arbitrator decides who uses the bus

25. Centralized-Parallel Arbitration

26. Daisy Chain Arbitration
Slow, but simple and cheap

27. Bus Summary