1. Networks Research Group Prof. Mark Handley Department of Computer Science

2. Nets Faculty: Interests & Accomplishments  PhDs from Cambridge, Harvard, MIT, Queen Mary, UCL  Internet standards activity (30+ standards including SIP)  Brought the first Internet link to Europe  Research contributions in: congestion control, distributed systems, mobility, multimedia, routing, scheduling, security, wireless networks.

3. Wedge: Making Networked Applications more Resilient to Attack Andrea Bittau, Petr Marchenko Mark Handley, and Brad Karp

4. Software vulnerability reports per year  As one-off defenses are released, new vulnerability categories emerge Source: Open source vulnerability database (osvdb.org)

5. Two systems security principles  Compartmentalization: Divide system into subsystems that fail independently  Least privilege: Each subsystem should only have access to read/modify data needed for its job

6. Problem: Lack of compartmentalization  Monolithic process must invoke SSL, so must hold RSA private key in memory  Single, monolithic address space: any code can access all memory Apache web server Internet request parser HACKED! private key

7. Wedge: Compartments improve security  Crowbar, a run-time instrumentation tool:  Measures memory access behavior of code running on real, non-malicious workloads  Informs programmer of memory permissions code requires on these workloads  OS primitives for Linux:  Fine-grained tagging of memory  Explicit assignment of per-tag permissions to threads request parser gen_session_key (~200 LoC ) memory protection Internet HACKED! private key

8. Cone of Silence: Nulling Interference for Higher-Throughput Wireless Networks Yiorgos Nikolaidis, Astrit Zhushi, Kyle Jamieson, and Brad Karp

9. The 802.11 (WiFi) “success disaster”

10. Limits of omnidirectional antennas  Interference from other senders is typically the limiting factor  Receiver can only decode Sender’s packet when Sender is sufficiently stronger than Interferer Interferer Receiver Sende r

11. The promise of directional antennas Interferer Receiver Sender OK!

12. Our approach: Cone of Silence (CoS)  Phased arrays: Long used in radar systems; newly available for 802.11  Receiver shapes beam to maximize SIR  Receiver balances between nulling toward interferer, directing gain toward sender  Explicitly measure S and I to account for multi-path propagation

13. Multipath TCP: Utilizing the natural resilience of the Internet Mark Handley, Damon Wischik, Costin Raiciu, Christoper Plunkte

14. Multipath traffic control, or, why peer-to-peer will balance the Internet Why does my phone have to choose between connections? What if it could use several connections at the same time?

15. Mb/s on wifi alone Mb/s on 3G alone Mb/s with multipath At my desk. Good wifi reception, poor 3G. Go downstairs to make coffee. Wifi fails. 3G is good. In the kitchen. Wifi is OK, 3G is good. We’ve implemented multipath extensions for TCP.

16. But is it safe?  If everyone greedily takes all the bandwidth they want, the Internet will collapse, like it did in 1988.  If everyone is restrained in the total bandwidth they use, but everyone greedily shifts all his/her traffic onto the best path, the Internet will “flap”.  Need to guarantee it is safe to deploy.

17. Not only is it safe, it’s the natural evolution of the Internet. Before the Internet, networks used to split links into “circuits”, e.g. one circuit per telephone call. One of the big ideas of the Internet was to use packets, not circuits, to carry data. This lets users take more bandwidth on a link as they need it — but relies on them to share fairly. Our algorithm extends the idea of “fair sharing as needed” from links to networks.

18. Build your own Internet  Trend: Opening the network up to innovation  Software routers, software-defined radios, peer-to-peer networks  Result: New ways of communicating