Transcript ppt
My CIDR Epiphany: Real World Data, Schema, and Environment Michael Franklin UC Berkeley Post SIGMOD PC Research Symposium (old persons track) February 11, 2005 Michael Franklin, UC Berkeley How it Happened or why it sometimes pays to hang around until the end of a conference • • • • The “gloom and doom” panel DeWitt’s gong show challenge Grappa consumption & staying up too late A great last session on sensor/stream processing, including: – Jennifer Widom’s Trio Talk – Shawn Jeffery’s HiFi Talk – Sam Madden’s Probabilistic Sensor Net Talk Michael Franklin, UC Berkeley The SIGMOD Credo Codd made relations, all else is the work of man. Leopold Kronecker (paraphrased by Raghu Ramakrishnan) Michael Franklin, UC Berkeley Database Management: Then Michael Franklin, UC Berkeley Database Management: Now Michael Franklin, UC Berkeley RM has been tremendously successful, but at a cost • Shoehorn the world into regular, flat tables. – This works particularly well for data that looks like regular, flat tables. • Ignore inconvenient facts about real world. – Source of a multi-billion $/yr consulting industry. • But, new applications, environments, devices, user expectations, are finally reaching a tipping point — stretching the model beyond its inherent capabilities. Michael Franklin, UC Berkeley Relational Model Assumptions: Real World Data All data in the database is 100% Valid The facts in the database are self-consistent Anything outside of the DB does not exist Time and space are just regular attributes Data items unambiguously map to real world entities Michael Franklin, UC Berkeley RM Assumptions: Schema All data conforms to a strict schema These schemas and their relationship to the data don't change much Everyone agrees on the meaning of the data No one cares where the data came from Michael Franklin, UC Berkeley RM Assumptions: Environment Users know exactly what they want to ask of the database Users want absolute answers (no satisficing) Queries can be independent of the user’s context All data is always available Michael Franklin, UC Berkeley Bridging the Physical Divide • We need to build systems that more realistically model the real world (and all its ambiguity) • We need to build systems that support users and conform to their goals, requirements, and habits (not vice versa) • This is going to require new data and query models, and likely another 30 years of work to get it right. Michael Franklin, UC Berkeley RM Assumption Cheat Sheet (A baker’s dozen) 1) 2) 3) 4) 5) All data in the database is 100% Valid The facts in the database are self-consistent Anything outside of the DB does not exist Time and space are just regular attribute Data items unambiguously map to real world entities Real World Data 6) 7) All data conforms to a strict schema These schemas and their relationship to the data don't change much Everyone agrees on the meaning of the data No one cares where the data came from Schema 8) 9) 10) Users know exactly what they want to ask of the database 11) Users want absolute answers (no satisficing) 12) Queries can be independent of the user’s context 13) All data is always available Michael Franklin, UC Berkeley Environment