Big Data is changing abruptly, and where it is likely heading
25 likes6,088 views
Big Data technologies are changing rapidly due to shifts in hardware, data types, and software frameworks. Incumbent Big Data technologies do not fully leverage newer hardware like multicore processors and large memory spaces, while newer open source projects like Spark have emerged to better utilize these resources. Containers, clouds, functional programming, databases, approximations, and notebooks represent significant trends in how Big Data is managed and analyzed at large scale.
Big Data is changing abruptly, and where it is likely heading
- 1. Big Data is changing abruptly, and where it is likely heading Big Data Spain 2014-11-17 bigdataspain.org Paco Nathan @pacoid
- 2. Some questions, and an emerging idea… The economics of datacenters has shifted toward warehouse scale, based on commodity hardware that includes multicore and large memory spaces. ! Incumbent technologies for Big Data do not embrace those changes, however newer OSS projects have emerged to leverage them. 2
- 3. Some questions, and an emerging idea… Also, the data requirements are changing abruptly as sensor data, microsatellites, and other IoT use cases boost data rates by orders of magnitude. ! We have math available to address those industrial needs, but have software frameworks kept up? 3
- 5. If you’re not using them, you’re missing out… ! See recent announcements by Google, Amazon, etc. ! Containers are the tip of the iceberg: OS have evolved quietly, while practices near to customers have perhaps not kept pace 5 Containers
- 6. Key benefits based on Mesos case studies: • mixed workloads • higher utilization rates • lower latency for data products • significantly lower Ops overhead ! Examples: Mesos @Twitter, eBay, Netflix, many startups, etc.; Borg/Omega @Google, Symphony @IBM, Autopilot @Microsoft, etc. ! Major cloud users understand this shift; however much of the world still thinks in terms of VMware 6 Containers
- 7. We need much more than Docker, we need better definitions for what is to be used outside of the container as well – e.g., for microservices • Intro to Mesos: goo.gl/PbUfNe • https://mesosphere.com/ • A project to watch: Weave 7 Containers
- 8. 8 Containers One problem: how does an institution, such as a major bank, navigate this kind of change?
- 9. 2. Clouds 9
- 10. Speaking of clouds, forgive me for being biased… For nearly a decade, I’ve been talking with people about whether cloud economics makes sense? Consider: datacenter util rates in single-digits (~8%) 10 Clouds
- 11. Companies accustomed to cheap energy rates, with staff who thinking in terms of individual VMs… these will not find clouds to be a friendly territory. Not so much about software, as it is about sophisticated engineering practice. The change recalls the shift of BI ⇒ Data Science That was not simple… 11 Clouds
- 12. The majors understand this new world: Google, Amazon, Microsoft, IBM, Apple, etc. Sub-majors can create their own clouds: Facebook, Twitter, LinkedIn, etc. 12 Clouds What about industries approaching Big Data now which cannot afford to buy small armies of expert Ops people? How they must proceed becomes a key question.
- 13. Current research points to significant changes ahead, where machine learning plays a key role in the context of advanced cluster scheduling: 13 Clouds Improving Resource Efficiency with Apache Mesos Christina Delimitrou youtu.be/YpmElyi94AA Experiences with Quasar+Mesos showed: • 88% apps get >95% performance • ~10% overprovisioning instead of 500% • up to 70% cluster util at steady state • 23% shorter scenario completion
- 14. 3. Abstract Algebra 14
- 15. A question on StackOverflow in 2013 challenged Twitter’s use of Abstract Algebra libraries for their revenue apps… cs.stackexchange.com/questions/9648/what-use-are-groups- monoids-and-rings-in-database-computations The answers are enlightening: think of a kind of containerization for business logic – leading to exponential increase in performance, particularly in the context of streaming cases. 15 Abstract Algebra
- 16. Abstract Algebra A cheat-sheet: non-empty set Semigroup has an identity element Group has a binary associative operation, with closure each element has an inverse has two binary associative operations: addition and multiplication Ring Monoid
- 17. Add ALL the Things: Abstract Algebra Meets Analytics infoq.com/presentations/abstract-algebra- analytics Avi Bryant, Strange Loop (2013) • grouping doesn’t matter (associativity) • ordering doesn’t matter (commutativity) • zeros get ignored In other words, while partitioning data at scale is quite difficult, you can let the math allow your code to be flexible at scale Avi Bryant @avibryant Abstract Algebra
- 18. Algebra for Analytics speakerdeck.com/johnynek/ algebra-for-analytics Oscar Boykin, Strata SC (2014) • “Associativity allows parallelism in reducing” by letting you put the () where you want • “Lack of associativity increases latency exponentially” Oscar Boykin @posco Abstract Algebra
- 19. 4. Functional Programming 19
- 20. Functional Programming for Big Data Theory, eight decades ago: what can be computed? Haskell Curry haskell.org Alonso Church wikipedia.org Praxis, four decades ago: algebra for applicative systems John Backus acm.org David Turner wikipedia.org Reality, two decades ago: machine data from web apps Pattie Maes MIT Media Lab 20
- 21. Functional Programming for Big Data circa 2002: mitigate risk of large distributed workloads lost due to disk failures on commodity hardware… Google File System Sanjay Ghemawat, Howard Gobioff, Shun-Tak Leung research.google.com/archive/gfs.html ! MapReduce: Simplified Data Processing on Large Clusters Jeffrey Dean, Sanjay Ghemawat research.google.com/archive/mapreduce.html 21
- 22. Functional Programming for Big Data DryadLINQ influenced a new class of workflow abstractions based on functional programming: Spark, Flink, Scalding, Scoobi, Scrunch, etc. • needed for leveraging the newer hardware: multicore, large memory spaces, etc. • significant code volume reduction ⇒ eng costs 22
- 23. Functional Programming for Big Data 2002 2004 MapReduce paper 2002 MapReduce @ Google 2004 2006 2008 2010 2012 2014 2006 Hadoop @ Yahoo! 2014 Apache Spark top-level 2010 Spark paper 2008 Hadoop Summit 23
- 24. Functional Programming for Big Data MapReduce Pregel Giraph Dremel Drill S4 Storm F1 MillWheel General Batch Processing Specialized Systems: Impala GraphLab iterative, interactive, streaming, graph, etc. Tez MR doesn’t compose well for large applications, and so specialized systems emerged as workarounds 24
- 25. Functional Programming for Big Data circa 2010: a unified engine for enterprise data workflows, based on commodity hardware a decade later… Spark: Cluster Computing with Working Sets Matei Zaharia, Mosharaf Chowdhury, Michael Franklin, Scott Shenker, Ion Stoica people.csail.mit.edu/matei/papers/2010/hotcloud_spark.pdf ! Resilient Distributed Datasets: A Fault-Tolerant Abstraction for In-Memory Cluster Computing Matei Zaharia, Mosharaf Chowdhury, Tathagata Das, Ankur Dave, Justin Ma, Murphy McCauley, Michael Franklin, Scott Shenker, Ion Stoica usenix.org/system/files/conference/nsdi12/nsdi12-final138.pdf 25
- 26. Functional Programming for Big Data One minor problem: where do we find many graduating students who are good at Scala, Clojure, F#, etc.? 26
- 27. 4. Databases 27
- 28. +1 generations of computer scientists has grown up thinking that Data implies SQL… Excellent as a kind of declarative, functional language for describing data transformations However, in the long run we should move away from thinking in terms of SQL Computational Thinking as a rubric in general would be preferred 28 Databases
- 29. A step further is to move away from thinking of data as an engineering issue – it is about business So much of the US curricula for math is still stuck in the Cold War: years of calculus to identify the best people to build ICBMs We need more business people thinking in terms of math, beyond calculus, to contend with difficult problems in supply chain, maintenance schedules, energy needs, transportation routes, etc. 29 Databases
- 30. Databases 1. starting with real-world data ⇒ 2. leverage graph queries for representation ⇒ 3. convert to sparse matrix, use FP and abstract algebra ⇒ 4. achieve high-ROI parallelism at scale, mostly about optimization ⇒
- 32. Approximations 19-20c. statistics emphasized defensibility in lieu of predictability, based on analytic variance and goodness-of-fit tests ! That approach inherently led toward a manner of computational thinking based on batch windows ! They missed a subtle point… 32
- 33. 21c. shift towards modeling based on probabilistic approximations: trade bounded errors for greatly reduced resource costs highlyscalable.wordpress.com/2012/05/01/ probabilistic-structures-web-analytics-data- mining/ 33 Approximations
- 34. 21c. shift towards modeling based on probabil approximations: trade bounded errors for greatly reduced resource costs Twitter catch-phrase: “Hash, don’t sample” highlyscalable.wordpress.com/2012/05/01/ probabilistic-structures-web-analytics-data- mining/ 34 Approximations
- 35. a fascinating and relatively new area, pioneered by relatively few people – e.g., Philippe Flajolet provides approximation, with error bounds – in general uses significantly less resources (RAM, CPU, etc.) many algorithms can be constructed from combinations of read and write monoids aggregate different ranges by composing hashes, instead of repeating full-queries 35 Approximations
- 36. algorithm use case example Count-Min Sketch frequency summaries code HyperLogLog set cardinality code Bloom Filter set membership MinHash set similarity DSQ streaming quantiles SkipList ordered sequence search 36 Approximations
- 37. algorithm use case example Count-Min Sketch frequency summaries code HyperLogLog set cardinality code suggestion: consider these as your most quintessential collections data types at scale Bloom Filter set membership MinHash set similarity DSQ streaming quantiles SkipList ordered sequence search 37 Approximations
- 38. • sketch algorithms: trade bounded errors for orders of magnitude less required resources, e.g., fit more complex apps in memory • multicore + large memory spaces (off heap) are increasing the resources per node in a cluster • containers allow for finer-grain allocation of cluster resources and multi-tenancy • monoids, etc.: guarantees of associativity within the code allow for more effective distributed computing, e.g., partial aggregates • less resources must be spent sorting/windowing data prior to working with a data set • real-time apps, which don’t have the luxury of anticipating data partitions, can respond quickly 38 Approximations
- 39. Probabilistic Data Structures for Web Analytics and Data Mining Ilya Katsov (2012-05-01) A collection of links for streaming algorithms and data structures Debasish Ghosh Aggregate Knowledge blog (now Neustar) Timon Karnezos, Matt Curcio, et al. Probabilistic Data Structures and Breaking Down Big Sequence Data C. Titus Brown, O'Reilly (2010-11-10) Algebird Avi Bryant, Oscar Boykin, et al. Twitter (2012) Mining of Massive Datasets Jure Leskovec, Anand Rajaraman, Jeff Ullman, Cambridge (2011) 39 Approximations
- 40. 7. Notebooks 40
- 41. Spreadsheets represented a revolution in how to think about computing ! What Fernando Perez, et al., have done with IPython is subtle and powerful ! Technology is not just about writing code; it is about people using that code ! Enterprise data workflows are moving toward cloud-based notebooks 41 Notebooks
- 42. Early emphasis on tools in Big Data now gives way to workflows: people, automation, process, integration, test, maintenance, scale Google had the brilliant approach of shared documents… Now we see cloud-based notebooks, which begin to displace web apps 42 Notebooks http://nbviewer.ipython.org/ http://jupyter.org/
- 44. What is Spark?
- 45. What is Spark? Developed in 2009 at UC Berkeley AMPLab, then open sourced in 2010, Spark has since become one of the largest OSS communities in big data, with over 200 contributors in 50+ organizations spark.apache.org “Organizations that are looking at big data challenges – including collection, ETL, storage, exploration and analytics – should consider Spark for its in-memory performance and the breadth of its model. It supports advanced analytics solutions on Hadoop clusters, including the iterative model required for machine learning and graph analysis.” Gartner, Advanced Analytics and Data Science (2014)
- 46. What is Spark?
- 47. What is Spark? Spark Core is the general execution engine for the Spark platform that other functionality is built atop: ! • in-memory computing capabilities deliver speed • general execution model supports wide variety of use cases • ease of development – native APIs in Java, Scala, Python (+ SQL, Clojure, R)
- 48. What is Spark? WordCount in 3 lines of Spark WordCount in 50+ lines of Java MR
- 49. What is Spark? Sustained exponential growth, as one of the most active Apache projects ohloh.net/orgs/apache
- 51. • generalized patterns ⇒ unified engine for many use cases • lazy evaluation of the lineage graph ⇒ reduces wait states, better pipelining • generational differences in hardware ⇒ off-heap use of large memory spaces • functional programming / ease of use ⇒ reduction in cost to maintain large apps • lower overhead for starting jobs • less expensive shuffles 51 What is Spark?
- 52. What is Spark? Kafka + Spark + Cassandra datastax.com/documentation/datastax_enterprise/4.5/ datastax_enterprise/spark/sparkIntro.html http://helenaedelson.com/?p=991 github.com/datastax/spark-cassandra-connector github.com/dibbhatt/kafka-spark-consumer unified compute data streams columnar key-value
- 53. databricks.com/blog/2014/11/05/spark-officially-sets- a-new-record-in-large-scale-sorting.html 53 What is Spark?
- 54. What is Spark? Databricks Cloud databricks.com/blog/2014/07/14/databricks-cloud-making- big-data-easy.html youtube.com/watch?v=dJQ5lV5Tldw#t=883
- 55. Resources 55
- 56. certification: Apache Spark developer certificate program • http://oreilly.com/go/sparkcert • defined by Spark experts @Databricks • assessed by O’Reilly Media • establishes the bar for Spark expertise 56
- 57. community: spark.apache.org/community.html video+slide archives: spark-summit.org events worldwide: goo.gl/2YqJZK resources: databricks.com/spark-training-resources workshops: databricks.com/spark-training new: MOOCs via edX and University of California 57
- 58. books: Fast Data Processing with Spark Holden Karau Packt (2013) shop.oreilly.com/product/ 9781782167068.do Spark in Action Chris Fregly Manning (2015*) sparkinaction.com/ Learning Spark Holden Karau, Andy Konwinski, Matei Zaharia O’Reilly (2015*) shop.oreilly.com/product/ 0636920028512.do 58
- 59. events: Strata EU Barcelona, Nov 19-21 strataconf.com/strataeu2014 Data Day Texas Austin, Jan 10 datadaytexas.com Strata CA San Jose, Feb 18-20 strataconf.com/strata2015 Spark Summit East NYC, Mar 18-19 spark-summit.org/east Strata EU London, May 5-7 strataconf.com/big-data-conference-uk-2015 Spark Summit 2015 SF, Jun 15-17 spark-summit.org 59
- 60. presenter: monthly newsletter for updates, events, conf summaries, etc.: liber118.com/pxn/ Just Enough Math O’Reilly, 2014 justenoughmath.com preview: youtu.be/TQ58cWgdCpA Enterprise Data Workflows with Cascading O’Reilly, 2013 shop.oreilly.com/product/ 0636920028536.do 60