Cost-Aware Traffic Management Under Demand Uncertainty from a Colocation Data Center User's Perspective

Burstable billing is widely adopted by colocation data center providers to charge their users for data transferring. This paper proposes a cost-aware traffic management approach for a colocation data center user under burstable billing where it is charged based on the 95th percentile bandwidth usage. To do this, we first develop a tractable mathematical expression to calculate the 95th percentile usage of a user. Then, we develop an optimization problem to maximize the user's surplus based on both deterministic and stochastic predictions of the user's demand. We show that the resulted optimization problem, while non-convex by nature, can be efficiently solved or approximated using a convex program. We also show that the proposed approach can also be applied in a more general scenario where the user gets services from multiple service providers. Using real-world workload traces, we show that the proposed approach can reduce a colocation data center user's IP transit cost by 26 percent and increase its total surplus by 23 percent, compared to the current practice of allocating bandwidth on-demand.

Automated summary

This paper introduces a cost-aware traffic management approach for colocation data center users under burstable billing, utilizing mathematical expressions to calculate 95th percentile usage and optimization problems to maximize user surplus. The resulting non-convex optimization problem can be efficiently solved or approximated using a convex program. Real-world workload traces show a reduction in IP transit cost by 26 percent and an increase in total surplus by 23 percent compared to current bandwidth allocation practices.