Efficiency

Source Routing
1st DFG/GENI
Doctoral Consortium,
San Juan, PR
March 13th-15th, 2011
Ashish Vulimiri, P. Brighten Godfrey
University of Illinois at Urbana-Champaign
Research Objectives
Abstract
• Goal: study efficiency and security
• Current Internet: inflexible
• Routing fixed within network
• Single offered path may be broken, congested, insecure
• Efficiency: can sources quickly discover good paths?
• Identify high-performance route selection algorithms
• Source routing: let packet specify path instead of destination
• Switch quickly in response to failures
• Use multiple routes simultaneously
• Choose application-appropriate service
• Security: source routing gives sources a lot of flexibility. Could a malicious
source (e.g. DoS attacker) misuse this flexibility?
• Study amount of damage a malicious source can cause
• Both realistic (Internet-like) traffic patterns and artificial worst-case
scenarios
• Question: how can sources use this flexibility? And can they misuse it?
Efficiency
Efficiency: Setup
• Want to optimize a specific metric, say
latency
Efficiency: Results
• Global PlanetLab overlay, up to 40 path choices per
source
• Source routing: network is dumb,
gives source no path status feedback
• Need to figure out path performance
from end-to-end probes
• Metrics:
• Optimality, convergence time,
overhead
Security
• DoS attacks
• Traditional solution: classify, separate
out malicious traffic. Problems:
• Hard to do well
• Efficiency
• Alternative: congestion pricing
• Network auctions off scarce
resources among competing users
• Source routing: users pick paths,
then pay at each edge in path
Future Work
• Efficiency: current work looks at
latency. Extend to bandwidth, loss
rate, jitter etc.
• Endhosts: PlanetLab or ProtoGENI
nodes
• Intermediate network: we tested
• Overlay [PlanetLab]
• In-network deployment [OpenFlow]
Security
• Game-theoretic analysis:
• Provable absolute efficiency guarantees
• ... as well as relative: can show pricing is
never worse, and frequently better, than
network-implemented fair queueing
• Current: simulations
• Future: implement congestion pricing in
network -- ProtoGENI
GENI Infrastructure
• Security: implement, study dynamics
• Have been using GENI infrastructure to test
against Internet-like background traffic
• Early implementation on PlanetLab
• Also a more recent setup on OpenFlow +
some PlanetLab/ProtoGENI nodes
• Compare overlays and in-network
deployments
• Should be of broader interest
(outside of the source-routing
contex)
• Currently planning experiments that need
provisioned links (with b/w and delay
guarantees)
• Planning to make more extensive use of
ProtoGENI
• Also tested on OpenFlow deployment
• Demoed a video streaming application at GEC9
Publications
• Two manuscripts in preparation:
• Efficiency (“Adaptive routing with end-toend feedback”)
• Security (“Can congestion pricing
neutralize DoS attacks?”)