Resource pricing and the evolution of congestion control

1
Resource pricing and the evolution of congestion
control

• By R. J. Gibbens and F. P. Kelly

2
A proportionally fair pricing.

• A fair distribution according to a price the user
  is willing to pay.
• Why ?
• How ?

3
Rates according to shadow pricing

• Let
• Then The change in the rate is

4
Rates according to shadow pricing

• If w(t) wr Then the stable point of the system
  is A proportionally fair per unit charge.

5
Congestion Mechanisms

• Creating various measurements and congestion
  control algorithms in the network itself
  (routers).floyd and fall
• Creating incentives for the end nodes to use
  congestion control charge aware TCP

6
Different approaches to charge aware TCP

• Paris metro pricing
• Smart market

7
The Expected Cost and Shadow price
8
The Expected Number of marks
9
When distribution is more general
Thus
10
Congestion Algorithm 1 the Elastic User(w)
Where
11
Congestion algorithm 2File Transfer(F,W)
Elastic User that changes the Payment.
12
Queue Marking Mechanism

• Problem Packets that arrive early at the busy
  period leave without being markedPackets that
  arrive after loss may be marked (although their
  shadow path is 0).

13
Queue Marking Mechanisms

• When a packet is lost mark all the packets in the
  queue and mark additional number.
• 1(Variant)Mark every packet from the first loss
  to the time the queue become empty

14
Queue - Marking Mechanisms (2)

• 2. Mark with probability calculated from the
  history of the queue.
• 3. Mark when ever a smaller virtual queue loses
  packet.

15
Comparison with the Internet Packet conversation
principle

• A new packet isnt put into the network until
  the old packet leaves self clocking

16
Solving the problems

• Slow-start exponential increase to the window
  size Increase with each ack received
• Congestion avoidance1. Additive increase.2.
  Multiplicative decrease.

17
Current congestion algorithm disadvantages

• Not user specific.
• Dropping packages is an extreme mechanism for
  congestion control.
• The rate at which the signals a generated in the
  source.

18
Response of end-nodes to Congestion

• Jacobson Average Rate
• Elastic user - Inverse proportion to

19
Jacobson Average Rate in our Equations

• If the user needs the average rate of Jacobson
  than the utility function would produce that
  rate.

20
Self Clocking in our Equations

• When no congestion indications are present
  File-transfer is doubling its rate (with
  proportion to T).

21
Self Clocking in our Equations

• Elastic User can be self clocking if cwnd
  increased by
• So the change in the rate is

22
Game Theory Model

• If the user is price-aware he will maximize
• The solution isWhen

23
Game Theory
The average paying is
When ?r? is constant and equal Then
Conclusion users shade their bids if they
have market power
24
Concluding remarks

• By appropriately marking the resources
  end-nodes are provided with the necessary
  information to make efficient use of the
  network resources