Trade & Cap: A customer-managed, market-based system for trading bandwidth allowances at a shared link

Jorge Londoño, Azer Bestavros, Nikolaos Laoutaris

TL;DR

This paper introduces "Trade & Cap" (T&C), a novel market-based system that allows users to voluntarily trade bandwidth allowances on a shared link. By enabling users to strategically acquire bandwidth slots, T&C incentivizes efficient resource utilization and leads to smoother traffic patterns, benefiting both users and Internet Service Providers (ISPs).

What Problem Does This Paper Address?

The ever-increasing demand for bandwidth-hungry applications like P2P, streaming, and Video on Demand (VoD) puts immense pressure on ISP infrastructures. Traditional flat-pricing schemes offer little incentive for users to moderate their bandwidth consumption, especially during peak hours, which are the primary drivers of ISP costs. Attempts by ISPs to implement per-byte pricing have been met with customer resistance. Furthermore, current congestion control mechanisms can be perceived as unfair, with low-volume users effectively subsidizing "bandwidth hogs." While ISPs have explored traffic shaping and differentiated service, these approaches often raise privacy concerns and can conflict with Net Neutrality principles. This paper addresses the need for a system that allows users to self-manage their bandwidth consumption in a fair and efficient manner, without requiring ISPs to act as intrusive arbiters.

Methods

The core of the Trade & Cap (T&C) system is a two-phase market-based mechanism:

  1. Bandwidth Trading Phase: Software agents, acting on behalf of users, engage in a non-cooperative game to strategically bid for and acquire bandwidth slots for their "Reserved Traffic" (RT). RT refers to applications requiring a specific minimum bandwidth over a contiguous period. The cost function is designed to incentivize users to shift their RT allocations away from congested time slots, promoting better utilization. This phase aims to reach a Nash Equilibrium where no agent can unilaterally improve its outcome.
  2. Bandwidth Capping Phase: After RT allocations are settled, this phase distributes any remaining capacity as "Fluid Traffic" (FT) to applications that benefit from opportunistic bandwidth. The allocation of FT is based on the remaining "buying power" of each user, rewarding those who preserved more budget in the trading phase. This phase ensures an aggregate cap on link utilization, often guided by the 95/5 rule, which is a common ISP metric for capacity planning.

The system is designed to be implemented with client-side agents (on user routers) and a provider-side agent (at the DSLAM), facilitating a distributed marketplace.

Datasets

The paper uses publicly available WAN traces for evaluation. The specific trace used is characterized by its total packets, TCP/UDP packet counts, and total TCP bytes. The authors note that direct DSLAM traces were not available for their study.

Key Insights and Ideas

  • User-driven fairness: T&C shifts the responsibility of defining "fairness" from the ISP to the users themselves, allowing them to collectively agree on resource allocation through a market mechanism.
  • Incentivizing flexibility: The concept of "slack" – the ability for users to shift their reserved bandwidth allocations in time – is crucial. Greater slack leads to more fluid traffic allocation and better overall utilization.
  • Smoothing traffic for ISPs: By encouraging users to trade and shift bandwidth, T&C significantly reduces the peak traffic volume, directly lowering ISP costs associated with infrastructure upgrades and traffic management.
  • "Net Neutrality" compliant: The ISP's role is reduced to that of an enforcer of the agreed-upon allocations, without needing to inspect or classify traffic content, thus adhering to Net Neutrality principles.
  • Dual nature of traffic: The system explicitly accounts for the difference between "Reserved Traffic" (requiring guarantees) and "Fluid Traffic" (opportunistic), a distinction often overlooked in simpler congestion control schemes.

Key Results

  • User benefits: Allowing users to schedule their RT requests with even a small amount of flexibility (e.g., 15 minutes of slack) can lead to a 16% reduction in their 95% traffic envelop. With 1 hour of slack, this reduction can reach up to 31%.
  • ISP benefits: The T&C system, especially with user flexibility, can reduce the ISP's 95th percentile traffic volume by up to 31%. More significantly, it dramatically smooths traffic volume, reducing the ratio of 95th-percentile to 50th-percentile traffic from 1.58 to nearly 1.004, effectively eliminating costly traffic spikes.
  • Fluid traffic increase: As user slack increases, the allocated fluid traffic also increases, demonstrating that the market mechanism effectively incentivizes users to reveal their traffic flexibility.
  • Equilibrium convergence: Analytical results establish that the pure strategies game in the trading phase converges to a Nash Equilibrium, and that an equilibrium always exists for the fluid bandwidth allocation phase.

Limitations and Future Work

  • Trace limitations: The authors acknowledge that the WAN traces used for evaluation might have a lower peak-to-valley ratio than typical real-world workloads, suggesting that the observed benefits might be even more significant in practice.
  • Implementation complexity: While a distributed architecture is proposed, the practical deployment and management of agents for potentially millions of users would require further consideration.
  • User adaptation: The paper assumes rational agents. Future work could explore how users might adapt their strategies or how the system could handle non-rational behavior.
  • Beyond DSLAM: The authors suggest that T&C is applicable to other shared resource management scenarios, such as aggregate inter-ISP traffic, which could be a direction for future research.

Summary updated

BibTeX
@inproceedings{londono2010trade,
  author = {Jorge Londoño and Azer Bestavros and Nikolaos Laoutaris},
  title = {{Trade \& Cap: A customer-managed, market-based system for trading bandwidth allowances at a shared link}},
  booktitle = {NetEcon},
  year = {2010},
  doi = {10.1145/1879082.1879092},
}
Figure