Application Layer I

Question 1

Communicating processes

Answer 1

Processes in different
hosts communicate by
exchanging messages

Messages go through
the protocol stack

Question 2

Header encapsulation/decapsulation

Answer 2

This is used to achieve
separation of concerns between each layer

Question 3

Identifying processes remotely

Answer 3

A process is addressed over the network using two
identifiers:
– IP address
* Network Layer identifier
* 32 bits (IPv4) or 64 bits (IPv6)– Port number
* Transport Layer identifier (16 bits)

– IP address is used by routers to forward messages to the correct host

– Then, the host’s OS uses the port number to forward messages to the
correct target process/ socket

Question 3

Sockets: The interface between Processes and Transport Protocols

Answer 3

The OS provides a so-called “socket” interface

Processes send/receive
messages to/from sockets

Client process: Initiates
communication

Server process: Waits to be contacted

Question 3

Port numbers

Answer 3

well known ports(common ones): range from 0- 1023
and
uncommon ones range from 1024-65535

Question 4

UNIX’s Socket API

Answer 4

In UNiX, almost every resource and device is made to look like a file

▪ A socket is like a file

• E.g., ‘standard’ API calls (“system calls”) are used - like read(), write(), close()
• Plus some other socket-specific calls: send(), recv(), bind(), etc

Question 5

Application layer protocols

Answer 5

Define types of messages: e.g., request, response, add-user.

Specify message structure:

Syntax: what fields are in a message.

Semantics: meaning of each field.

Control communication rules:

When and how messages are sent and responded to.

Types:

Open protocols: Publicly defined (e.g., HTTP, SMTP), ensure compatibility between systems.

Proprietary protocols: Private, product-specific (e.g., Skype, AppleTalk).

Question 6

What might an application need from
a transport service?

Answer 6

Data integrity
– Some apps require 100%
reliable data transfer (e.g.
file transfer, web
transactions)
– Other apps (e.g., audio) can tolerate some loss

Throughput(data per second)
- Some apps require a
minimum level of throughput
(e.g. multimedia)
- Other apps can make use of whatever throughput they get (“elastic apps”)

Timing– Some apps (e.g., Internet
telephony, interactive
games) require low delay

Security- Encryption, data
integrity

Question 7

Internet transport protocols

Answer 7

TCP service:

• Reliable transport between sending and
  receiving process
• All messages are delivered in the
  order they were sent
• Flow control: sender can’t overwhelm
  receiver

Connection-oriented: initial setup is required
between client and server processes

• Does not provide: timeliness, minimum
  throughput guarantee, security

UDP service:

Unreliable data transfer between
sending and receiving process

• Does not provide: reliability, in-order
  delivery, flow control, congestion
  control, timeliness, throughput
  guarantee, security, or connection
  setup

BUT timeliness may be better than
TCP; and less overhead may be
incurred for both hosts and the
network

Question 8

Socket programming with TCP

Answer 8

TCP service is connection-oriented:
– When a client application creates a socket, the client’s
TCP instance establishes a connection to the server’s TCP instance

When contacted by a client, a server TCP creates a new
socket for server process to communicate with that specific
client

Question 9

Raw Sockets

Answer 9

Commonly used for implementing custom network protocols and
capturing incoming ICMP packets

• ICMP packets do not contain a transport layer
• Therefore, they do not include port numbers
• This means ICMP packets are not sent to a process

By setting the value of the protocol as socket.IPPROTO_ICMP, you can capture all the ICMP traffic on your host

• Typically requires elevated privileges (administrator/root) to create
  raw sockets due to security concerns

Question 10

Client-Server architecture

Answer 10

Server:
–An always-on host
–Has permanent IP address
–(Can use data centres for scaling)

• Clients:
  – system is not always connected to the network
  –Private/ephemeral IP addresses
  –Do not communicate directly
  with each other

Question 11

P2P architecture(s)

Answer 11

No always-on central server.

Devices (peers) talk to each other directly.

Each peer can request and provide services (like sharing files).

Scales well: new peers bring more demand and capacity.

Peers may be online only sometimes and their IP addresses change, making it harder to manage.