Spark!
6 likes1,802 views
This document discusses Spark, an open-source cluster computing framework. It notes that while Hadoop is useful for batch processing, it has limitations for interactive and iterative algorithms. Spark addresses these issues through its resilient distributed datasets (RDDs) which can be operated on in parallel and rebuilt if lost. RDDs support transformations like map and filter as well as actions that return values. The document provides examples of using Spark from Scala and discusses its architecture involving a DAG scheduler and task scheduler.
![HOW TO USE IT?
scala> val textFile = sc.textFile("README.md")
textFile: spark.RDD[String] = spark.MappedRDD@2ee9b6e3
!
scala> textFile.count() // Number of items in this RDD
res0: Long = 74
!
scala> textFile.first() // First item in this RDD
res1: String = # Apache Spark
!
scala> textFile.map(line => line.split(" ").size).reduce((a, b) =>
Math.max(a, b)) // How many words are in the longest line
res2: Int = 16
!
scala> textFile.flatMap(line => line.split(" ")).map(word => (word,
1)).reduceByKey((a, b) => a + b).collect
res3: Array[(java.lang.String, Int)] = Array((need,2), ("",43), (Extra,
3), (using,1), (passed,1), (etc.,1), (its,1), (`/usr/local/lib/
libmesos.so`,1), (`SCALA_HOME`,1), (option,1), (these,1), (#,1),
(`PATH`,,2), (200,1), (To,3),...](https://image.slidesharecdn.com/spark-140221045158-phpapp01/85/Spark-13-320.jpg)
Spark!
- 2. HTTP://ABOUT.ME/PRZEMEK.MACIOLEK/ • Data Scientist, Hadoop user since 2009 • Did research for Academia, data mined for oil&gas exploration industry, cofounded Data Science startup, built Big Data team in Base CRM, … • A lot of different tools used meanwhile (Mahout, HBase, Cassandra, Redis, Pig, Storm, …) • Dreaming about something powerful and concise for Big Data… • AD 2014: Head of Analytics & Data @ Toptal - researching new ways of doing Big Data Analytics, rediscovered Storm. P.S. Ever considered doing Analytics & Data Science for a very cool startup? Drop me a note at: [email protected]
- 4. HADOOP IS COOL (BUT SOMETIMES IT’S NOT) • High latency (interactive, anyone?) • Challenging expressibility of business logic • Iterative algorithms? (think: PageRank)
- 8. MAYBE MAP REDUCE IS NOT ALWAYS THE BEST SOLUTION?
- 9. GENERALIZE FTW! Spark Task DAG and Data Sharing MapReduce … Batch processing Specialized systems
- 10. RESILIENT DISTRIBUTED DATASET (RDD) • A collection of elements that can be operated in parallel • Parallel Collection, e.g. sc.paralellize(Array(1,2,3)) • Hadoop Dataset • Lazily evaluated, able to rebuild lost data any time • Can be stored in memory without replication
- 11. ACTIONS TRANSFORMATIONS • Creates a new dataset from an existing one • • Return the value to the driver after computation finishes • Runs all required transformations Lazily evaluated • Recomputed each time an action runs on it, but might be persisted (in memory or disk) • Broadcast Variables and Accumulators for cluster-level sharing
- 13. HOW TO USE IT? scala> val textFile = sc.textFile("README.md") textFile: spark.RDD[String] = spark.MappedRDD@2ee9b6e3 ! scala> textFile.count() // Number of items in this RDD res0: Long = 74 ! scala> textFile.first() // First item in this RDD res1: String = # Apache Spark ! scala> textFile.map(line => line.split(" ").size).reduce((a, b) => Math.max(a, b)) // How many words are in the longest line res2: Int = 16 ! scala> textFile.flatMap(line => line.split(" ")).map(word => (word, 1)).reduceByKey((a, b) => a + b).collect res3: Array[(java.lang.String, Int)] = Array((need,2), ("",43), (Extra, 3), (using,1), (passed,1), (etc.,1), (its,1), (`/usr/local/lib/ libmesos.so`,1), (`SCALA_HOME`,1), (option,1), (these,1), (#,1), (`PATH`,,2), (200,1), (To,3),...
- 15. RDD Objects DAG Scheduler Split graph into stages of tasks. Submit each one when ready. rdd.filter().map(…). groupBy(…).filter(…) t Se sk Ta Worker Execute tasks. Store and serve blocks. Task Task Scheduler Lunch tasks via cluster manager. Retry.
- 16. NARROW DEPENDENCIES WIDE (SHUFFLE) DEPENDENCIES map, filter groupByKey union join (inputs not co-partitioned)
- 18. How much code is needed to implement Big Data Page Rank?
- 23. BERKELEY DATA ANALYTICS STACK * https://amplab.cs.berkeley.edu/software/
- 24. SPARK LIVE