Auction or T

1
Auction or Tâtonnement Finding Congestion
Prices for Adaptive Applications

• Xin Wang Henning Schulzrinne
• Columbia University

2
Outline

• Motivations
• Goal and scope
• Pricing strategies
• Congestion pricing
• Performance studies
• Conclusion

3
Motivations

• Two types of congestion pricing models in
  Internet today
• Tâtonnement
• Iteratively update price that aggregate user
  demand approaches available bandwidth
• Auction
• Allocate bandwidth based on users bidding price
• No work comparing their performance, and little
  work address the practical issues

4
Goal and Scope

• Compare tâtonnement and auction
• Develop comparable pricing models
• Compare the performances
• Address issues for practical usages
• Scope
• Network
• periodical price adjustments/resource allocations
• Users
• short-term reservation/demand adaptations

5
Pricing Strategies

• Holding price and charge
• based on cost of blocking other users by holding
  bandwidth even without sending data
• Usage price and charge
• maximize the providers profit, constrained by
  resource availability
• Congestion price and charge
• drive demand to supply level (tâtonnement or
  auction)

6
Congestion Pricing

• Tâtonnement process
• Congestion charge proportional to excess demand
  relative to target utilization
• M-bid auction model
• User indicates its willingness to pay a premium
  for different bandwidths under congestion through
  bids
• Congestion price charge highest rejected bid
  price
• Features
• reduce uncertainty user can express multiple
  preferences
• reduce signaling bursts user provides bids in
  advance
• reduce setup delay inter-auction admission
  allowed
• support periodical auctions
  

7
Resource Allocations

• Tâtonnement
• User agent determines the optimal demand based
  on user preferences, network price, constrained
  by user budget and application QoS requirements.
• Auction
• Network selects bids exceeding the auction
  price multiple bids of a user can be higher
  than the auction price, select the one with
  higher bandwidth (lower price per unit
  bandwidth).
• User agents adapt rate based on allocated
  bandwidth/QoS from network auctions.

8
Simulation Topologies
Topology 1
Bottlenecks Studied
Topology 2
9
Tâtonnement vs. auction
Tâtonnement can maintain the target utilization
(0.9) With similar user benefit, auction has
higher utilization Blocking of tâtonnement is 40
times smaller than that of fixed pricing
blocking of auction is almost zero, with
fractions of users delayed until next auction.
10
Tâtonnement vs. auction (contd)
Tâtonnement and auction have comparable total and
average user benefit Tâtonnement has higher
congestion price, and hence allows for higher
network revenue.
11
Impact of target utilization and M
Higher throughput of tâtonnement is at the cost
of performances higher blocking probability and
lower user benefit Auction performance is robust
to the variations of M
12
Performance of topology-2
Similar trends as those of single bottleneck
13
Conclusions

• Performance comparisons
• Tâtonnement and auction
• effectively control congestions
• have comparable performances
• function effectively over a range of parameters
• control periods, demand elasticity, different
  numbers of user multiplexing, different network
  topologies.
• Auction has higher bandwidth utilization at a
  given user benefit, but has
• higher implementation complexity
• longer setup delay
• Tâtonnement has higher network revenue
• Resolve some practical issues for both schemes