Internet and Media Distribution

1
Internet and Media Distribution

• By Jeremiah Scholl

2
Agenda

• Intro to computer networks
• A bit of Internet History
• A bit of Internet architecture
• Some design limitations
• Strategies for dealing with those limitations

3
What is a computer network?

• A network of computers ?
• A collection of computers that can share
  information (data) with each other over a network.

4
Building a computer network

• Some engineering issues.
• How should we connect the computers together?

5
Ways of connecting computers

• Communicate Directly
• Mesh Connect all computers together. Reliable
  and efficient but expensive
• Bus All computers share a common wire and
  broadcast on the wire to reach each other.
• Computers other than the one you want to reach
  ignore the broadcast

6
Ways of connecting computers

• Communicate Indirectly
• routers and end-hosts
• Ring Computers are connected
• in ring with data flowing in one direction.
• Cheaper than mesh, but less reliable
• Star A central computer acts as
• a go-between.

7
Some Internet History
8
Problem

• 1960s the US military wanted to build a large
  computer network
• Computer networks were small and/or centrally
  controlled.
• Cold war mentality
• What if we get nuked?
• Cities will burn
• People will die
• Our network will go down! (if they hit the
  central command)

9
A centrally controlled network

• Attack the control center to take out the whole
  network

10
A centrally controlled network

• Attack the control center to take out the whole
  network

11
1969 ARPANET

• The goal was to build a decentralized scalable
  network.
• Sections must be able to operate independently if
  other parts of the network are destroyed.
• First generation ARPANET connected 5 hosts in
  1969.

12
How was this done?

• By creating a network of networks
  (internetwork).
• Each network can grow and die independently.

13
An example internetwork
14
Each subnet works on its own
15
How was this accomplished?

• Postal system is a metaphor can describe the
  Internet
• A packet switched network.
• A universal addressing system is used (i.e.
  130.240.66.68)
• Data is packaged up into small pieces and sent
  on the network.
• Packets contain sender and receiver addresses
  etc.
• Goal to keep routers as stateless (stupid) as
  possible.
• Routers have enough information to forward
  packets.
• Routers only know about their local neighborhood
• And know how reach other computers by sending
  packets in the right direction
• Routers have little other information.

16
Birth of the Internet?

• Stupid routers require lots of intelligence to
  be pushed into end-hosts.
• Standards for the operation of routers and
  end-hosts were needed.
• A family of communication protocols called
  TCP/IP was standardized in 1982.
• IP is implemented in routers and TCP in
  end-hosts.
• The Internet back-bone has changed little since
  then.

17
Some limitations of the Internet
18
An alternative network

• Telephone networks do not work like the Internet
• Circuit-switched instead of packet switched.
• Reserves reserved throughout the entire network
  during a call.
• The network assumes it knows what you are sending
  (voice).
• Performs services like echo-cancellation.

19
Remember

• Routers on the Internet do NOT
• make assumptions about what kind of data they are
  delivering.
• keep other state information.
• like who is communicating with whom

20
Some Limitations

• This means that routers can not
• know how much data each computer is sending.
• know if some data is more important than others
• know if I want to send data to more than 1 person
  

21
Congestion

• One of the more serious problems created by
  stupid routers.
• Occurs if too many computers want to send data at
  the same time.

22
Congestion

• Packets enter a queue and wait before being sent.
• The longer the queue the
• longer the wait.
• Eventually the queue
• fills up.
• Packets start getting dropped!!!

23
Best-effort service

• For this reason the Internet is said to provide
  best-effort service
• No guarantees about long it will take to deliver
  a packet
• No guarantees the packet will get there at all
• I wont even be informed if the packet is dropped!

24
Adding Reliability

• Transport Control Protocol (TCP) runs at end
  hosts (sender and receiver).
• Handles packet recovery in case of loss.
• Receivers send an ACK (I got it) back to the
  sender when a packet arrives.
• Senders sends a repair if no ACK is received.
• This process takes time.
• Estimates the amount of available bandwidth.
• It is used for the vast majority of traditional
  Internet traffic.
• Email, FTP, The Web

25
What about new media?

• TCP works well for traditional media.
• New media like audio and video have strict time
  requirements.
• TV/Movies are normally around 25 fps ( 1 frame
  every 0.04)
• Best-effort service leads to variability in time
  delivery.
• TCP makes this even worse.
• Just use a playout buffer ?
• No good for interactive media (like conferencing)
  ?

26
A Proactive Approach

• Sending repairs creates delay.
• Why not just send multiple copies of the data?
• If one copy is lost we are still ok
• Less delay but more overhead
• This is called Forward Error Correction
• Used commonly for audio and sometimes for video

27
FEC (Forward Error Correction)
28
Interleaving
29
One other limitation

• The Internet does not provide multicast service.
• This makes it hard to reach a very large
  audience.

30
Unicast (bad) vs Multicast (good)
31
Reaching large groups

• Have end hosts replicate data and pass it to each
  other
• Use dedicated servers
• I.E. Web Caching
• Use end user machines
• Like in file sharing

32
The future?

• The last 15 years has seen TONS of research into
  solving these problems.
• Diffserve allows important data to be given
  priority.
• IP-Multicast allows
• routers to replicate data.
• Both are used in subnets today.