Presentation Layer, Congestion Control

• Lossy Compression
• Security
• Congestion Control
• Queueing Disciplines
• Fairness

Presentation Layer: Lossy Compression

• JPEG:
  • discrete cosine transform (almost lossless)
  • quantization: reduced precision of high spatial frequencies
  • run-length encoding of quantized data
• MPEG:
  • JPEG on each frame
  • 3 frame types:
    • I (Intrapicture), like JPEG
    • P (Predicted), specified in terms of the differences from prior I frame
    • B (Bidirectional Predicted), specified in terms of the differences from prior I/P frame to next I/P frame

Presentation Layer: Security

• trusted system(s): own computer, partner's computer
• untrusted system: network
• plaintext encrypted to give cyphertext, decrypted to give back plaintext
• secret key is required for decryption, may be required for encryption:
  • if encryption and decryption keys are same, encryption key must be kept secret: this is secret key cryptography
  • if encryption and decryption keys are same, encryption key may be published: this is public key cryptography
  • in both cases, the decryption key {\bf must} be kept secret to guarantee privacy

Security Algorithms, Protocols

• one-time pad: unbreakable, but key security is difficult (secret key)
• DES: hard to break 56-bit key, commercial use (secret key)
• Kerberos: security protocol, distributes and renews session keys based on host keys (secret key)
• RSA: arbitrary size keys, public keys
• message integrity: can I prove in court that this message was sent to me by person X?

In-Class Discussion

• how would you make contact with a perfect stranger over an insecure medium (e.g. the internet) using:
  • public-key encryption
  • private-key encryption
• what attacks are possible?
• how do you know who you are talking with? does it matter?
• how does this change if you are allowed one secure exchange before you start?

Congestion Control

• April 6th:
  • flow control slows down the sender the receiver's pace
  • congestion control slows down the sender to avoid overwhelming the network:
    • finite router buffer space means discarding packets when buffers full
    • even with infinite buffer space, data would get stale as it sits in the queues
• discarding packets wastes resources, makes network unpredictable
• who is "causing" the congestion?
  • everyone
  • those who increase their sending rate

Network Model

• packet switched network
• connectionless, but routers keep soft state (" flows" -- see IPv6 flow identifier)
• best-effort vs. QoS
• who is responsible for congestion control?
  • routers: have more information
  • hosts: have more control
• reservation or feedback
• window-based or rate-based

Fairness

• absolute fairness: everyone gets the same bandwidth out of the network
• capitalism: everyone gets bandwidth proportional to what they pay
• socialism: everyone gets bandwidth proportional to what they need
• local fairness: every flow gets the same bandwidth at each router (parking structure problem)
• QoS: you don't get the bandwidth you need, you get the bandwidth you reserved

Router Queueing Strategies

• FIFO:
  • packets are added to outbound queue in the order received
  • packets received while queue is full are discarded
  • end-systems become responsible for congestion control
• Fair Queueing:
  • each flow has a logical queue
  • packets are sent in order one from each non-empty queue
  • packets from the fastest flows are discarded first
  • enforces local fairness

Fair Queueing with Variable-Sized Packets

• for variable-sized packets, must compute fair time to send
• record for each packet, as it arrives, stamp F_i, where
  • F_i = max(F[i-1], A_i)+P_i
  • F[i-1] is the stamp of the packet before us in this flow
  • A_i is the actual time of the packet arrival
  • P_i is the packet size (in time to transmit)
• always send the packet with the lowest value of F