110 likes | 202 Views
BubbleStorm Overview. Christof Leng, Max Lehn, Robert Rehner. Peer-to-Peer: A Failure?. The hype is over cornered in the file-sharing niche few new applications appear most applications are relatively primitive research results are too complex for practical use What developers want:
E N D
BubbleStorm Overview Christof Leng, Max Lehn, Robert Rehner 18.04.2011 | Databases and Distributed Systems
Peer-to-Peer: A Failure? • The hype is over • cornered in the file-sharing niche • few new applications appear • most applications are relatively primitive • research results are too complex for practical use • What developers want: • powerful queries (keyword search, SQL, XPath) • proper abstraction • predictable reliability • easy to use 18.04.2011 | Databases and Distributed Systems
Separation of Concerns Expressive search is easy Frontend Application Logic Backend Data Store BubbleStorm Middleware BubbleStorm Middleware Black Box Internet 18.04.2011 | Databases and Distributed Systems
Powerful search Push and Pull possible O(√n) replicas Very scalable Network size static Query/Data ratio static No resilience against churn Principle of Rendezvous Search search request Web Server query new page found Crawler data 18.04.2011 | Databases and Distributed Systems
CAP-Theorem (also known as Brewer‘s theorem) You can never satisfy all three of the following guarantees Consistency • all nodes see the same data at the same time Availability • node failures do not prevent survivors from continuing to operate Partition tolerance • the system continues to operate despite arbitrary message loss Wegiveupsomeconsistency andtrytomaximizethetwootherguarantees 18.04.2011 | Databases and Distributed Systems
O(√λn) replicas Probabilistic guarantees Flexible balancing Resilient against churn BubbleStorm Concept Searching Node query Publisher Node data BubbleStorm: Resilient, Probabilistic, and Exhaustive Peer-to-Peer SearchWesley W. Terpstra, Jussi Kangasharju, Christof Leng, Alejandro P. BuchmannProceedings of the 2007 ACM SIGCOMM Conference, Kyoto, Japan, August 2007 18.04.2011 | Databases and Distributed Systems
Random Topology 18.04.2011 | Databases and Distributed Systems
Bubblecast 18.04.2011 | Databases and Distributed Systems
Three Types of Replication Fading Data • fire & forget • no updates • cannot be deleted • no overhead • will diminish over time Managed Data • tied to a maintainer • updates / deletes serialized by maintainer • new nodes poll for managed data • disappears when maintainer leaves • orphaned / inconsistent data gets flushed eventually Durable Data • should stay forever (unless deleted) • use unstructured key-based routing • update / delete possible • eventual consistency enabled by churn Maintaining Replicas in Unstructured P2P SystemsChristof Leng, Wesley W. Terpstra, Bettina Kemme, Wilhelm Stannat, Alejandro P. BuchmannProceedings of ACM CoNEXT 2008, Madrid, Spain, December 2008 Publication under preparation 18.04.2011 | Databases and Distributed Systems
BubbleStorm Wrap-up • Resilient, probabilistic peer-to-peer search overlay • Tailor-made for complex queries or publish/subscribe • Mutable metadata stored in the network • Different replication modes for typical scenarios • Easy to use for application developers 18.04.2011 | Databases and Distributed Systems
Thanks for listening! ? Questions http://www.bubblestorm.net http://www.dvs.tu-darmstadt.de 18.04.2011 | Databases and Distributed Systems