PeopleNet: An Internet for the people






PeopleNet Overview

In the Internet of the future, people will find it more intuitive to both exchange and retrieve multimedia based content.

Multimedia traffic is poised to account for more than 90% of all consumer Internet traffic. A variety of multimedia applications consistently demand loss-free, low-latency paths from the Internet. However, Internet protocols and policies, are highly optimized for elastic applications like text/images. While overlay networks promise application specific routing that often outperform Internet path selection, current overlay architectures either require extensive path probing or apriori knowledge to select alternate (superior) paths.

PeopleNet investigates ways to build a large-scale multimedia overlay on top of the present day Internet that can consistently provide loss-free, latency reducing alternate paths to multimedia applications; without the need for background monitoring or any apriori knowledge. Using Internet measurements from 550+ vantage points in the Internet for over a week, we show that it is sufficient to periodically attempt a random subset of k nodes in any overlay of N nodes, where k is bounded by O(ln(N)), to derive loss-free, low-latency paths. Such path selection strategies often outperform Internet path selection for multimedia applications, and can scale to service millions of nodes. We investigate various properties of such random subsets to derive simple and powerful guidelines to build a generic Internet service that a variety of multimedia applications can query.

Related Publications