|
CSE Home | CSE 588 Home | Scribe List | About Us | Search | Contact Info |
Naming Systems DNS what are the biggest problems in DNS insertion capability failure context sensitivity generality / mechanism (Akami, Napster, Distributed Director) politics name allocation - haphazard time (cost) for missing name failover time "Poison a DNS cache" authoritative server may be secure, but proxies not attacker can trigger lookup, and send reply as through were authoritative server - because no authentication between authoritative server and proxies fix?? IPsec - if control protocol preload data security - probably biggest issue for most Freenet Chord LDAP Is peer to peer file sharing something that a naming system should support? probably beter to leave naming system simple, and put other functionality on top of it. Possibly still use DNS, and build on top of (different from Freenet and Chord). what applications of naming? features goals performance fast lookup fast update search scalability clients / servers large # hosts # names # lookups updates to hosts and names reliable accuracy availability Simplicity low layer simple - build functionality on top vs. implement rich functionality into product ubiquity portability security authentication of updates only authorized updates / restricted? plausible deniability / secrecy Access control for lookup management overhead delegation no centralization context sensitive naming flexible (aliases) many-to-many mapping BoA --} server A bank of American --} server B .... API lookup context-based (wildcard) on keyword, not data do we want to support webcrawl, google type searches? example: want to find all printers in Seattle. DNS doesn't support easily does naming service become a generic DB - with SQL-like support reverse lookup triggers (eg notification when change occurs) Applications IP lookup updates Reverse Lookups non-hierarchical naming? Object location is something like Google a solution to the file sharing problem? put files on webserver, let google do search, build the context-based data How build system DNS (example, as in book, with hierarchy of servers) how make scalable? caching - every time go to root server, cache partial result .edu washington.edu cs.washington.edu www.cs.washington.edu (actual webpage) long timeouts invalidate (digression on cache coherency) directory-based invalidation cache coherence use TTL on data replicate (root servers) - possibly with single IP addr partition data among root servers (or hash) update just need to change table at relevant level parallel servers to zone transfers - some latency on update could system be modified to support searching? not easily. need to go through entire tree to find data Akamai returns different results from root server searches based on source of request (points to local proxy) long TTL at root server (hours), short TTL at proxy (seconds) how build geographic list of IP addresses? do you care about geography, rather than network distance most routers have location as part of name
Department of Computer Science & Engineering University of Washington Box 352350 Seattle, WA 98195-2350 (206) 543-1695 voice, (206) 543-2969 FAX [comments to owner-cse588] |