130 likes | 247 Views
The Design and Implementation of a Next Generation Name Service for the Internet. V. Ramasubramanian, E. Gun Sirer Cornell Univ. SIGCOMM 2004. Ciprian Tutu – Systems Seminar 8/4/04 Johns Hopkins University. DNS: Current Operation and Issues.
E N D
The Design and Implementation of a Next Generation Name Service for the Internet V. Ramasubramanian, E. Gun Sirer Cornell Univ. SIGCOMM 2004 Ciprian Tutu – Systems Seminar 8/4/04 Johns Hopkins University
DNS: Current Operation and Issues • High latency in query resolve (low cache hit-rates) • High load on root and TLD servers • Slow update propagation (40% have TTL > 1 day) • Lame delegations • Implementation errors (?)
CoDoNS Goals • High Performance • Low latency, increased lookup performance • Resilience to Attacks • Decentralization • Dynamic load balancing • Fast Update Propagation • Support secure delegation
Beehive • Prefix-matching DHT • O(logN) lookup • Pastry, Tapestry • Proactive caching • O(1) lookup • C=0.5 hops • xi=fraction of objects replicated at level I • b=DHT base
CoDoNS: Architecture • Decouples namespace management from query resolution • Domain names mapped to 128bit unique identifiers • Direct caching for locality • Home node stores permanent copies of RR’s • No TTL associated with records inside CoDoNS • Supports negative caching (NXDOMAIN)
CoDoNS (cont.) • Supports DNSSEC signatures • Caches certificates • Insert/Update use version number to prevent replay attacks. (!! not Dynamic DNS compliant) • Allows multiple operators to manage the same part of the name hierarchy • If conflicting records, clients “simply” pick records signed by an operator they trust (?!) • CoDoNS uses its own centralized authority to sign resource records fetched from legacy DNS (!!)
CoDoNS Evaluation • MIT trace • 12 hours; 281,943 queries; 47,230 unique domain names • Deployed on 75 PlanetLab nodes Query Resolution Latency
CoDoNS: Flash-crowd Effect Avg bw: 12.2KB/s/node AvgRecords/node: 4217 (10% of total, 13MB storage)
CoDoNS: Update Propagation • For 1 million node CoDoNS network it would take less than 1 minute to update 99% of replicas
Conclusions • Decouple management from query resolution • Reduce resolver latency • Improve update propagation delay • Reduce load on root servers • Resistent to flash-crowd effect (?) • Attempt to eliminate monopoly in namespace management
Questions/Issues • Compatibility with dynamic DNS • Giving RR signing authority to CoDoNS • Not really great behaviour for flash-crowds • CoDoNS caches any data that is queried (size issues) • Selective caching? • No TTL on CoDoNS nodes -> if home node becomes partitioned, then no expiration. • Further issues related to CoDoNS network partitioning • Is there enough incentive for cooperation?