Seminar | April 20 | 1-2 p.m. | 400 Cory Hall
Carlee Joe-Wong, CMU
Recent growth in mobile data usage has been driven by increasing demand for an ever broader variety of mobile application content. Faced with these developments, Internet service providers (ISPs) are beginning to involve content providers (CPs) in their pricing and management of mobile data. However, this involvement risks disrupting the principle of network neutrality i.e., that all data traffic should be treated the same, regardless of its content. In this talk, I will consider this problem from two sides. On the one hand, CPs may be allowed to subsidize their users data consumption: this pricing scheme, called sponsored data, does not impact CPs or users quality-of-service (QoS). I will examine models for CPs optimal data sponsorship and show that, while sponsorship disproportionately benefits wealthier, cost-insensitive CPs, it benefits users more than CPs and disproportionately benefits cost-sensitive users. I will then turn to the scenario where ISPs might provide different QoS levels for different CPs traffic in exchange for a fee. Emerging software-defined networking technologies allow ISPs to provide such differentiated QoS by passing traffic through cloud servers that virtualize their division into different QoS levels. However, the volatility of data traffic can make it difficult to provision enough cloud resources to handle this traffic, affecting ISPs cost of handling data. Cloud service providers face similar problems in managing computational jobs. I will present some of our existing work on the feasibility of multiplexing short-lived computational jobs, in particular the tradeoff of cost savings with increased job completion times. I will finally suggest some ways to adapt these cloud models for the case of ISPs differentiating CP data traffic, causal dependency between sequences using the directed information.