Turning Heterogeneity into an Advantage in Overlay Routing
Explore advanced routing strategies for DHT overlays integrating expressway nodes and autonomous systems to enhance scalability and latency management. Compare content location methods for peer-to-peer systems, featuring interest-based locality shortcuts and optimized bandwidth-demanding techniques.
Turning Heterogeneity into an Advantage in Overlay Routing
E N D
Presentation Transcript
Turning Heterogeneity into an Advantage in Overlay Routing Published in INFOCOM 2003 Authors: Ahichen Xu(HP), Mallik Mahalingam(VMware), Magnus Karlsson(HP) Gisik Kwon Dept. of Computer Science and Engineering Arizona State University
Motivation • Exploiting physically efficient routing and peer heterogeneity over DHT-based overlay network • Constructing an auxiliary network • expressway
Default overlay : CAN and eCAN • Each node knows its neighbors in the d-space • Forward query to the neighbor that is closest to the query id • Example: assume n1 queries f4 7 6 n5 n4 n3 f4 5 4 f1 3 n2 n1 2 f3 1 f2 0 0 2 3 4 6 7 5 1
Brocade Architecture Brocade Layer Original Route Brocade Route AS-3 AS-1 S R AS-2 P2P Network
Expressway • Expressway nodes(EN) & expressway neighbors • Autonomous System(AS) topology • Landmark clustering • Route summary • Propagated periodically • All the local nodes in same AS
Routing Expressway node Ordinary node
Experiment • Stretch • The ratio of accumulated latency in the actual routing path to the shortest-path latency from the source to destination • Two topology • Internet-like topology derived from BGP report • Transit-stub graph by GT-ITM • Logical auxiliary • Brocade-like system
Comparison various approaches AS topology Transit-stub
TTL and Number of ENs Transit-stub AS topology
Efficient Content Location Using Interest-Based Locality in Peer-to-Peer Systems Published in INFOCOM 2003 Authors: Kunwadee Sripanidkulchai, Bruce Maggs, Hui Zhang (CMU) Excerpt from Kunwadee Sripanidkulchai’s presentatin file Gisik Kwon Dept. of Computer Science and Engineering Arizona State University
Motivation • Design goals • Decentralized • Simple and robust • Scalable • Let’s retain the simplicity and robustness of Gnutella and make it scalable • Locality! • Network locality? No. • Popularity? No. • Interest-based locality? Yes.
Interest-based locality Someone in my research group Random person on the street “If a peer has a particular piece of content that I am interested in, it is very likely that it will have other pieces of content that I am (will be) interested in as well.” 2002 Infocom proceedings? 2001 Infocom proceedings?
Our solution: Shortcuts • Overlay on top of Gnutella • Benefits • Can be easily integrated into Gnutella • Can be used with many other underlying mechanisms like DHT’s
Shortcut Where is ? Discover and add shortcut. Discover interest-based shortcuts No shortcut.
Where is ? Use interest-based shortcuts Shortcut Use shortcut. Success! O(1) scope for most searches. No index (state) maintained.
Constructing shortcuts • Shortcut discovery • Infer locality using underlying protocol (Gnutella) • Add 1 shortcut to list at a time • Shortcut selection • Rank shortcuts based on performance • Ask shortcuts sequentially • Limit shortcut list size to 10
Removing practical limitations • Shortcut discovery • Add 1 shortcut to list at a time • => add all peers returned from search • => discover shortcut through our existing shortcuts • Shortcut selection • Limit shortcut list size to 10 • => no bound
Measurement-Based Optimization Techniques for Bandwidth-Demanding Peer-to-Peer Systems Published in INFOCOM 2003 Authors: T.S.Eugene Ng, Yang-hua Chu, Sanjay G. Rao, Kunwadee Sripanidkulchai, Hui Zhang Gisik Kwon Dept. of Computer Science and Engineering Arizona State University
Motivation • Improve the performance with light-weight measurement-based techniques • Qualitative analysis • RTT probing • Smallest response to 36B ICMP ping message • 10KB TCP probing • Fastest download of 10KB data • Bottleneck bandwidth probing(BNBW) • Largest nettimer • Nettimer is a project to do end-to-end network performance measurement. • It can listen passively to existing network traffic or actively probe the network.
Performance metrics • Media file sharing • Optimality Ratio (OR) • The ration between the TCP bandwidth achieved by downloading from the selected server peer and the TCP bandwidth achievable from the best server peer in the candidate set • Overlay multicast streaming • Convergence time • The amount of time after the initial join it takes for the peer to receive more than 95% of the stable bandwidth for 30 seconds • stable bandwidth is determined based on the bandwidth it receives at the end of a 5-minutes experiment
Accuracy of choices 36B RTT 10KB TCP BNBW
Average OR UIUC CMU 10Mbps
Average OR U of Alberta CMU ADSL
Media file sharing • Joint ranking
Overlay multicast streaming • RTT • Single packet RTT probing • RTT filter + 10K • At most 5 best RTT -> 10KB downloading • RTT filter + 1-bit BNBW • At most 5 best RTT -> highest bottleneck BW
Convergence time Basic techniques Combined techniques