How often have you contemplated using a carrier pigeon as an adjunct to the piece of wet string that apparently carries data from your site to its counterparts? Cho and Gupta explore the concept in some detail and develop a planning system called Pandora that can be used to ensure that data reaches its destination(s) by a nominated deadline, at minimal cost.
They present a scenario where data is to be copied from a couple of university sites to the Amazon EC2 facility. The data can be moved entirely via the Internet or shipped on removable disk devices to the Amazon site via FedEx or a similar carrier service. Alternatively, one might choose to copy data from one university to the other on the Internet, and then aggregate the data there for subsequent FedEx transfer to the Amazon site.
The lowest-cost scheme will depend on the quantities of data at each site, the costs of transporting it using each mechanism, Amazon’s import fee, and the time(s) at which the data must be completely copied into the Amazon facility.
The data transfer problem is modeled using a network flow graph, where each link has capacity, cost, and transit-time attributes. A conversion is performed so that the problem can be solved using mixed integer programming techniques. Experiments were conducted using bandwidth figures from PlanetLab, shipping data from FedEx, and published Amazon import costs.
There are some interesting references relating to earlier systems (such as Postmanet) that explored the use of mixed-mode data delivery systems. If you have ever contemplated the use of carrier pigeons, you will be fascinated by this paper.