BIG DATA PROCESSING A DEEP DIVE IN HADOOP/SPARK & AZURE SQL DW

Size: px

Start display at page:

Download "BIG DATA PROCESSING A DEEP DIVE IN HADOOP/SPARK & AZURE SQL DW"

Regina Parker
6 years ago
Views:

1 BIG DATA PROCESSING A DEEP DIVE IN HADOOP/SPARK & AZURE SQL DW

2 TOPICS COVERED 1 2 Fundamentals of Big Data Platforms Major Big Data Tools

3 Scaling Up vs. Out SCALE UP (SMP) SCALE OUT (MPP) + (n) Upgrade components or buy bigger server each time Multiprocessor system where processors share resources : Operating System (OS) Memory I/O devices connected using a common bus Add nodes to the cluster Multiple processing nodes OS RAM Network

5 mins over 910 nodes Fastest sort of a TB, 62 secs over 1,460 nodes Sorted a PB in 16.

4 Innovation Timeline Doug Cutting & Mike Cafarella started working on Nutch Doug Cutting adds DFS & MapReduce support to Nutch NY Times converts 4TB of image archives over 100 EC2s Fastest sort of a TB, 3.5 mins over 910 nodes Fastest sort of a TB, 62 secs over 1,460 nodes Sorted a PB in hours over 3,658 nodes Google publishes GFS & MapReduce papers Yahoo! Hires Cutting, Hadoop spins out of Nutch Founded Doug Cutting joins Cloudera Facebooks launches Hive: SQL Support for Hadoop Hadoop Summit 2009, 750 attendees

5 THE FUNDAMENTALS OF HADOOP Hadoop evolved directly from commodity scientific supercomputing clusters developed in the 1990s Hadoop consists of: MapReduce Hadoop Distributed File System (HDFS)

6 WHAT S NEW

7 BASICS OF MPP 400 bills 1 bill/ sec = 400 Seconds

8 BASICS OF MPP 200 bills 1 bill/ sec = 200 Seconds 200 bills 1 bill/ sec = 200 Seconds Total = 200 Seconds

9 BASICS OF MPP 100 Bills 1 bill/ sec = 100 Seconds 100 Bills 1 bill/ sec = 100 Seconds 100 Bills 1 bill/ sec = 100 Seconds 100 Bills 1 bill/ sec = 100 Seconds Total = 100 Seconds

10 HDFS & MAPREDUCE The Main Node: runs the Job tracker and the name node controls the files. Each node runs two processes: Task Tracker and Data Node Map Reduce Job Tracker Task Tracker Task Tracker HDFS Cluster Name Node Data Node Data Node 1 N

11 BASICS OF MAPREDUCE The Main Node: runs the Job tracker and the name node controls the files. Each node runs two processes: Task Tracker and Data Node Query Result Data Nodes/Task Trackers Query Name Node/ Job Tracker

12 EXECUTION UNITS MAPREDUCE

13 SOME DISTRIBUTIONS OF APACHE HADOOP Apache Foundation

14 Sandbox Hortonworks

15 MAPREDUCE PIG & HIVE MAPREDUCE PIG HIVE Java Write many lines of code Mostly used by Yahoo Most used for data processing Shares some constructs w/ SQL Is more Verbose Needs a lot of training for users with limited procedural programming background Offers control over the flow of data Mostly used by Facebook for analytic purposes Used for analytics Relatively easier for developers w/ SQL experience Less control over optimization of data flows compared to Pig Not as efficient as MapReduce Higher productivity for data scientists and developers

16 THE EXPLOSION OF HADOOP

17 THE HISTORY OF SPARK MapReduce Top Level Spark Paper BSD Open Source Apache 17

18 SPARK SHARED LIBRARIES

19 SPARK THE UNIFIED PLATFORM FOR BIG DATA APIs for : Scala Java Python R Spark SQL Spark Streaming MLlib (machine learning) GraphX (graph) Spark Core

20 SPARK BENEFITS Performance Using in-memory computing, Spark is considerably faster than Hadoop (100x in some tests). Can be used for batch and realtime data processing. Unified Engine Integrated framework includes higher-level libraries for interactive SQL queries, processing streaming data, machine learning and graph processing. A single application can combine all types of processing. Developer Productivity Easy-to-use APIs for processing large datasets. Includes 100+ operators for transforming. Ecosystem Spark has built-in support for many data sources such as HDFS, RDBMS, S3, Apache Hive, Cassandra and MongoDB. Runs on top of the Apache YARN resource manager.

21 ANALYTICS CORTANA

22 SQL Server Big Data Optimizations

23 SQL Server APS

24 SQL Server APS Growth Topology Scale Unit Base Unit Base Unit Extension

25 SQL Server Azure SQL DW

26 SQL Server Deployment options and hybrid solutions

27 SQL Server Connecting Islands of Data with PolyBase Select Result set Provides a single T-SQL query model for PDW and Hadoop with rich features of T-SQL, including joins without ETL Microsoft Azure HDInsight Hortonworks for Windows and Linux Cloudera SQL Server Parallel Data Warehouse PolyBase Microsoft HDInsight Uses the power of MPP to enhance query execution performance Supports Windows Azure HDInsight to enable new hybrid cloud scenarios Provides the ability to query non- Microsoft Hadoop distributions, such as Hortonworks and Cloudera

28 USE CASE: SUPPLY CHAIN MANAGEMENT Use Case 3: Supply Chain Management

29 USE CASE: SMART GRID MANAGEMENT

30 USE CASES SMART GRID PREDICTIVE MAINTENANCE DEMAND FORECASTING GRID OPTIMIZATION THEFT PERVENTION DEMAND RESPONSE CUSTOMER PROFILING

32 USE CASES SMART GRID

33 USE CASE: REAL TIME TRAFFIC ANALYSIS

34 REAL TIME TRAFFIC ANALYSIS

35 USE CASES STREAM ANALYTICS Real-time fraud detection Connected cars Click-stream analysis Real-time financial portfolio alerts Smart grid, energy management CRM alerting sales to customer case Data and identity protection services Real-time sales tracking

36 ML PROBLEMS SOLVED BY AZURE ML Classification Regression Recommenders Anomaly Detection Clustering

37 INDUSTRY USE CASES MACHINE LEARNING

38 THE MACHINE LEARNING WORKFLOW Input data Data transformation Define model Split data Train model Score (prediction) Evaluate Model

39 AZURE DATA FACTORY

40 HD INSIGHT

41 BIG DATA & Advanced Analytics Roadshow Questions? Orion Gebremedhin

Analytics Platform System

Analytics Platform System Big data. Small data. All data. Audie Wright, DW & Big Data Specialist Audie.Wright@Microsoft.com Ofc 425-538-0044, Cell 303-324-2860 Sean Mikha, DW & Big Data Architect semikha@microsoft.com