![3 skills: data plumbing
Glue languages: Python, Perl, regex, XSLT
Admin: setting up, maintaining clusters
Af鍖nity with OSS & *nix
NoSQL = NoSchema = Transform Data
/^([w!#$%&'*+-/=?^`{|}~]+.)*[w!#$%&
'*+-/=?^`{|}~]+@((((([a-z0-9]{1}[a-z0-9-]{0,62}[a-
z0-9]{1})|[a-z]).)+[a-z]{2,6})|(d{1,3}.){3}d{1,3}(:d{1,5})?)$/i
Tuesday, June 8, 2010](https://image.slidesharecdn.com/bigdatabodenseebarcamp-100608035009-phpapp01/85/Big-Data-Bodensee-Barcamp-2010-14-320.jpg)
Big Data @ Bodensee Barcamp 2010
- 1. Tuesday, June 8, 2010
- 2. Tuesday, June 8, 2010
- 3. BIG DATA The rise of the data scientist http://鍖owingdata.com/2009/06/04/rise-of-the-data-scientist/ Tuesday, June 8, 2010
- 4. Holidaycheck Travel platform: review + book 12+ countries (.de ... .cn) 30% growth / year, profitable Almost 1.5 mio hotel reviews 1.6 mio + pics Tuesday, June 8, 2010
- 5. Data @ HC internet-driven 15 Gb Operational company Data traditional: MVC/ 12 Gb logs / day 3-Tier/RDBMS/ caching 5 searches / second 50+ Apache instances My scientist friend: Thats neat, but its not data science. Tuesday, June 8, 2010
- 6. The I/O Bottleneck The problem is simple: Memory, Disk size and CPU and even network performance continue to grow much faster than disk I/O performance. 2004 to 2009 CPU: still following Moore's Law (transistor x2 every 18 months) Memory Bandwidth (Intel): 9.3x Disk Density (SATA): 8x Disk I/O: 0.8x Network speed: routers can easily saturate the fastest hard drives http://blogs.cisco.com/datacenter/comments/networking_delivering_more_by_exceeding_the_law_of_moore/ Tuesday, June 8, 2010
- 7. I/O Repercussions Turn to memcache Try out SSD Try out asynchronous writes (e.g. message queues) Try to solve/hack the I/O problem: Sharding, in-memory DB Our problems seem big, but are they really? Tuesday, June 8, 2010
- 8. So what is Big Data anyway? The term Big data from software engineering and computer science describes datasets that grow so large that they become awkward to work with using on-hand database management tools kilo to mega to giga to tera to peta to exa to zetta to yotta Tuesday, June 8, 2010
- 9. NoSQL = Not Only SQL Trade-Offs, e.g. transactions, data loss e.g. Document Stores (MongoDB) e.g. Key-Value Stores (MemcacheDB) e.g. Graph Databases (Neo4j) Map/Reduce algorithm Tuesday, June 8, 2010
- 10. Medium Data With yesterday's scienti鍖c technology most businesses should be able to handle their data analysis needs. HC: 12 Gb log鍖les / day = medium data problem Solved (?) with: RDBMS + NoSQL (2006) Bigtable: A Distributed Storage System for Structured Data, Fay Chang, Jeffrey Dean, Sanjay Ghemawat, Wilson C. Hsieh, Deborah A. Wallach, Mike Burrows, Tushar Chandra, Andrew Fikes, and Robert E. Gruber (2004) MapReduce: Simpli鍖ed Data Processing on Large Clusters, Jeffrey Dean and Sanjay Ghemawat Tuesday, June 8, 2010
- 11. 3 sexy skills of data geeks The sexy job in the next ten years will be statisticians The ability to take datato be able to understand it, to process it, to extract value from it, to visualize it, to communicate it. Hal Valerian (Google) http://dataspora.com/blog/sexy-data-geeks/ Tuesday, June 8, 2010
- 12. 3 skills: statistics sentiment analysis machine learning natural language processing recommendation engines good old-fashioned regression Tuesday, June 8, 2010
- 13. 3 skills: visualization Q: Are you hiring statisticians, visualization experts & data plumbers? Vs. TheOathMeal Edward Tufte, Ben Fry Tuesday, June 8, 2010
- 14. 3 skills: data plumbing Glue languages: Python, Perl, regex, XSLT Admin: setting up, maintaining clusters Af鍖nity with OSS & *nix NoSQL = NoSchema = Transform Data /^([w!#$%&'*+-/=?^`{|}~]+.)*[w!#$%& '*+-/=?^`{|}~]+@((((([a-z0-9]{1}[a-z0-9-]{0,62}[a- z0-9]{1})|[a-z]).)+[a-z]{2,6})|(d{1,3}.){3}d{1,3}(:d{1,5})?)$/i Tuesday, June 8, 2010
- 15. More Data beats smart algorithms face recognition spelling correction machine translation http://videos.syntience.com/ai-meetups/peternorvig.html http://dataspora.com/blog/tipping-points-and-big-data/ Tuesday, June 8, 2010
- 16. Ethics of data Black Hat vs. White Hat <=> Black Data vs. White data White: Amazon free public datasets (e.g. human genome) Black: Scientific climate data (or the lack of PUBLIC data) Just like money, information flows to the least taxed location in a global world. Tuesday, June 8, 2010
- 17. Take-Away & Discuss Don't throw away data if you dont have to, because unlike material goods, data becomes more valuable the more of it is created. As a society, I don't think we understand this completely yet. q: Who is using a NoSQL db? Share Stories? q: Do you know how much data you are q: Do you hire statisticians? throwing away? q: Do you hire visualization q: Any tips on introducing NoSQL in experts? companies? q: Share: how big is your data? q: Do you own your customer data or q: Do you own your analytics data? does Facebook? q: How are you exploiting q: Do you own your content or does asynchronicity? Google? q: Should information be regulated (privacy)? Can it? Tuesday, June 8, 2010
Editor's Notes
- #2: Does a 500 Gb stick exist? yes, this is a quiz, internet is allowed no cheating, no SSD drives
- #3: Not it doesnt. Chinese fake. A bit better than this one. When will you think a 1 Tb USB stick will exist? Petabyte? We mostly believe in Moores law & thats a problem.
- #4: Big Data: what is it? Setup the systems. Data scientists: who are they? Hire the people. Discuss!
- #5: growing pains
- #7: The web is full of &quot;data-driven apps.&quot; We are one. But that does not make us data scientistsStorage & Analysis are separate things. : Operational vs. Analysis datastore
- #8: When designing systems, these days you run more and more into I/O bottlenecks.
- #9: NoSQL: document-stores, Turn in your schema at the entrance, trade-offs, MongoDB, Cassandra, NoSQL = Not ONLY SQL clickpaths question: describe data sizes in audience
- #10: Used to be: Big Oil. Big Telco. Big Banking. Big Pharma. BIG in Physics: LHC outputs 24 zettabytes / second. BIG in Genetics: several terabytes per sequencing experiment. Personal genome / Personalized medicine / less than 10 years ago human genome, now 1000 genomes project, SNPs (23andme) 10 & 24 zeroes Illumina sequencer /
- #12: yesterday = BigTable, MapReduce, Clustering approx. 5 years old Let's face it: most businesses do not have the data needs ... Exceptions: Google / Facebook / Twitter. Take away: can you handle medium-data? What tech can be used? What kind of systems can I build? NoSQL.
- #13: The human factor: who do I hire? http://radar.oreilly.com/2010/06/what-is-data-science.html http://dataspora.com/blog/sexy-data-geeks/ Do you have a st atistician on board? Do you have a data vi sualization expert on board?Do you have a data plumber on board?
- #14: When all of the above fails: crowdsourcing? MTurk
- #15: Edward Tufte, Ben Fry Do you have a statistician on board? Do you have a data visualization expert on board?Do you have a data plumber on board?
- #17: Peter Norvig spelling corrector, machine translation, image recognition Phase shifts: dig out data that you thought didnt exist: GayDar, Netflix
- #18: Project Gaydar: do you own yourself? Netflix competition: shreddingGoogle trading floor: buy more google stock!# Grey data23andMe:
- #19: Is that your data, or are you just happy to see me? How big is your data (Share)Who is using a NoSQL db? Share?Do you have statisticians? Visual experts? Data plumbe