80 likes | 192 Views
Architectural Considerations for Protecting End Hosts. Vern Paxson International Computer Science Institute and Lawrence Berkeley National Laboratory vern@icir.org June 28, 2007. Overview. Previous session looked at architectural issues for the network securing its own infrastructure
E N D
Architectural Considerations for Protecting End Hosts Vern Paxson International Computer Science Institute and Lawrence Berkeley National Laboratory vern@icir.org June 28, 2007
Overview • Previous session looked at architectural issues for the network securing its own infrastructure • Now, we consider the network’s role (if any) in protecting end systems • Two parts: • What should its role be? • Architectural approaches for DoS defense
Agenda, Part 1 • What should the network’s role be? • It’s inevitable that the network will be intrusive with end system traffic, let’s architect for it (Vern Paxson) • No let’s not, #1 (Paul Syverson) • No let’s not, #2 (Nick Weaver) • Discussion
Agenda, Part 2 • How should the network protect against DoS? • Framing of problem space (Stefan Savage) • The role of identifiers (Stefan Savage) • The role of indirection (Angelos Keromytis) • The role of capabilities (Xiaowei Yang) • Discussion
A Glum Vision ThatWe Had Better Plan For • Network operators want to control traffic • Control = inspect, modify, tune, censor, confine • … for a large variety of reasons: • Policy enforcement • Endhost security • Wiretap / legal intercept • Censorship • Walled gardens / business reasons • Performance engineering
Glum Vision, con’t • Furthermore, they have the power to do so since they hold the fundamental property of connectivity … • … unless they are constrained: • By law • Unpredictably difficult to shape in a useful fashion • By competitive concerns • But these are trumped by law and sole-sourcing
Glum Vision, con’t • We have existence proofs that network operators will go to significant lengths to shoehorn in such control • Question: would we like this stuff shoehorned into our future Internet, or directly recognized as a tussle? • (Q: will end-to-end crypto save us? A: No, reduces to steganography.)
How to Architect for this Tussle? • One approach (w/ S. Shenker, M. Allman): • When instantiating communication, end nodes negotiate with network regarding degree of inspection/meddling • What will be revealed, what can be modified • How to express range of semantics? • If unacceptable, seek alternate paths • Both parties require mechanisms to police for adherence • Already today for SIP proxies, P3P