PDL People

David Andersen

www |
GHC 9109
(412) 268-3064
(412) 268-5576

Gates-Hillman 9215
(412) 268-5099
Mailing Address: School of Computer Science
Carnegie Mellon University
5000 Forbes Avenue
Pittsburgh, PA 15213-3891
Associate Professor, CSD

Research Interests:

My research focuses on networks and distributed systems, with an eye towards improving the availability and performance of Internet-based systems.  My previous research includes the development of overlay networks to route around Internet failures, work on secure operating systems and storage systems, and Internet measurement.

Highly Available Internet Architecture
This project is examining new architectural concepts to improve the availability and security of the Internet, without compromising the fundamental flexibility that underlies its success.  The primitives we are currently exploring include enhancing end-hosts' ability to select between paths through the Internet, and permitting hosts or networks greater control over what traffic they receive, in a way that is enforceable deep inside the network.  This project emphasizes real-world measurements from our own Internet testbed as well as public testbeds such as Planetlab to understand the problems  facing today's network, and to evaluate potential solutions.

The Data-Oriented Transfer project, in collaboration with researchers at Intel Research, is examining a new way to structure Internet applications that perform bulk transfers.  Instead of performing the transfers themselves, these applications pass their data to a transfer service that performs the transfer on their behalf.  The transfer service serves as a locus for the development and deployment of novel transfer techniques.  Our initial efforts include merging e-mail delivery with peer to peer techniques (e.g., using peer to peer to collaboratively deliver large file attachments to multiple receivers), and developing transfer techniques to improve the performance of applications when the underlying network layers perform poorly.

Opportunistic Resource Use in Wireless Networks
is examining better ways to make use of wireless networks by explicitly exploiting concurrent multi-path transfers, by opportunistically caching overheard traffic, and by taking advantage of quiescent periods to preemptively transfer data across the network.  This project is attempting to turn one of wireless networks weaknesses—a shared broadcast medium—Into a strength, by taking advantage of the processing and storage power available at nodes to avoid expensive wireless communication.