Upgrade to Pro — share decks privately, control downloads, hide ads and more …

The Future of Analytics, Data Integration and B...

Mark RIttman
September 12, 2016

The Future of Analytics, Data Integration and BI on Big Data PlatformS

As presented in Dublin at the HUG IRL meetup

Mark RIttman

September 12, 2016
Tweet

More Decks by Mark RIttman

Other Decks in Technology

Transcript

  1. Mark Rittman, Oracle ACE Director THE FUTURE OF ANALYTICS, DATA

    INTEGRATION AND BI ON BIG DATA PLATFORMS HADOOP USER GROUP IRELAND (HUG IRL) Dublin, September 2016
  2. •Mark Rittman, Co-Founder of Rittman Mead •Oracle ACE Director, specialising

    in Oracle BI&DW •14 Years Experience with Oracle Technology •Regular columnist for Oracle Magazine •Author of two Oracle Press Oracle BI books •Oracle Business Intelligence Developers Guide •Oracle Exalytics Revealed •Writer for Rittman Mead Blog :
 http://www.rittmanmead.com/blog •Email : [email protected] •Twitter : @markrittman About the Speaker 2
  3. 4

  4. •Started back in 1996 on a bank Oracle DW project

    •Our tools were Oracle 7.3.4, SQL*Plus, PL/SQL 
 and shell scripts •Went on to use Oracle Developer/2000 and Designer/2000 •Our initial users queried the DW using SQL*Plus •And later on, we rolled-out Discoverer/2000 to everyone else •And life was fun… 20 Years in Old-school BI & Data Warehousing 6
  5. •Data warehouses provided a unified view of the business •Single

    place to store key data and metrics •Joined-up view of the business •Aggregates and conformed dimensions •ETL routines to load, cleanse and conform data •BI tools for simple, guided access to information •Tabular data access using SQL-generating tools •Drill paths, hierarchies, facts, attributes •Fast access to pre-computed aggregates •Packaged BI for fast-start ERP analytics Data Warehouses and Enterprise BI Tools 7 Oracle MongoDB Oracle Sybase IBM DB/2 MS SQL MS SQL Server Core ERP Platform Retail Banking Call Center E-Commerce CRM 
 Business Intelligence Tools 
 Data Warehouse Access &
 Performance
 Layer ODS /
 Foundation
 Layer 7
  6. •Examples were Crystal Reports, Oracle Reports, Cognos Impromptu, Business Objects

    •Report written against carefully-curated BI dataset, or directly connecting to ERP/CRM •Adding data from external sources, or other RDBMSs,
 was difficult and involved IT resources •Report-writing was a skilled job •High ongoing cost for maintenance and changes •Little scope for analysis, predictive modeling •Often user frustration and pace of delivery Reporting Back Then… 8 8
  7. •For example Oracle OBIEE, SAP Business Objects, IBM Cognos •Full-featured,

    IT-orientated enterprise BI platforms •Metadata layers, integrated security, web delivery •Pre-build ERP metadata layers, dashboards + reports •Federated queries across multiple sources •Single version of the truth across the enterprise •Mobile, web dashboards, alerts, published reports •Integration with SOA and web services Then Came Enterprise BI Tools 10 10
  8. •Data from new-world applications is not like historic data •Typically

    comes in non-tabular form •JSON, log files, key/value pairs •Users often want it speculatively •Haven’t thought it through •Schema can evolve •Or maybe there isn’t one •But the end-users want it now •Not when you’re ready But Why Hadoop? Reason #1 - Flexible Storage 16 Big Data Management Platform Discovery & Development Labs Safe & secure Discovery and Development environment Data sets and samples Models and programs Single Customer View Enriched Customer Profile Correlating Modeling Machine Learning Scoring Schema-on Read Analysis
  9. •Enterprise High-End RDBMSs such as Oracle can scale •Clustering for

    single-instance DBs can scale to >PB •Exadata scales further by offloading queries to storage •Sharded databases (e.g. Netezza) can scale further •But cost (and complexity) become limiting factors •Typically $1m/node is not uncommon But Why Hadoop? Reason #2 - Massive Scalability 17
  10. •Hadoop started by being synonymous with MapReduce, and Java coding

    •But YARN (Yet another Resource Negotiator) broke this dependency •Modern Hadoop platforms provide overall cluster resource management,
 but support multiple processing frameworks •General-purpose (e.g. MapReduce) •Graph processing •Machine Learning •Real-Time Processing (Spark Streaming, Storm) •Even the Hadoop resource management framework
 can be swapped out •Apache Mesos Reason #3 - Processing Frameworks 18 Big Data Platform - All Running Natively Under Hadoop YARN (Cluster Resource Management) Batch
 (MapReduce) HDFS (Cluster Filesystem holding raw data) Interactive
 (Impala, Drill,
 Tez, Presto) Streaming +
 In-Memory
 (Spark, Storm) Graph + Search
 (Solr, Giraph) Enriched 
 Customer Profile Modeling Scoring
  11. •Data now landed in Hadoop clusters, NoSQL databases and Cloud

    Storage •Flexible data storage platform with cheap storage, flexible schema support + compute •Data lands in the data lake or reservoir in raw form, then minimally processed •Data then accessed directly by “data scientists”, or processed further into DW Meet the New Data Warehouse : The “Data Lake” 19 Data Transfer Data Access Data Factory Data Reservoir Business Intelligence Tools Hadoop Platform File Based Integration Stream Based Integration Data streams Discovery & Development Labs Safe & secure Discovery and Development environment Data sets and samples Models and programs Marketing / Sales Applications Models Machine Learning Segments Operational Data Transactions Customer Master ata Unstructured Data Voice + Chat Transcripts ETL Based Integration Raw Customer Data Data stored in the original format (usually files) such as SS7, ASN.1, JSON etc. Mapped Customer Data Data sets produced by mapping and transforming raw data
  12. •Enterprise High-End RDBMSs such as Oracle can scale into the

    petabytes, using clustering •Sharded databases (e.g. Netezza) can scale further but with complexity / single workload trade-offs •Hadoop was designed from outside for massive horizontal scalability - using cheap hardware •Anticipates hardware failure and makes multiple copies of data as protection •More nodes you add, more stable it becomes •And at a fraction of the cost of traditional
 RDBMS platforms Hadoop : The Default Platform Today for Analytics 23
  13. •Cloudera’s answer to Hive query response time issues •MPP SQL

    query engine running on Hadoop, bypasses MapReduce for direct data access •Mostly in-memory, but spills to disk if required •Uses Hive metastore to access Hive table metadata •Similar SQL dialect to Hive - not as rich though and no support for Hive SerDes, storage handlers etc Cloudera Impala - Fast, MPP-style Access to Hadoop Data 39
  14. •Beginners usually store data in HDFS using text file formats

    (CSV) but these have limitations •Apache AVRO often used for general-purpose processing •Splitability, schema evolution, in-built metadata, support for block compression •Parquet now commonly used with Impala due to column-orientated storage •Mirrors work in RDBMS world around column-store •Only return (project) the columns you require across a wide table Parquet - Column-Orientated Storage for Analytics 40
  15. •But Parquet (and HDFS) have significant limitation for real-time analytics

    applications •Append-only orientation, focus on column-store 
 makes streaming ingestion harder •Cloudera Kudu aims to combine 
 best of HDFS + HBase •Real-time analytics-optimised •Supports updates to data •Fast ingestion of data •Accessed using SQL-style tables
 and get/put/update/delete API Cloudera Kudu - Best of HBase and Column-Store 41
  16. •Kudu storage used with Impala - create tables using Kudu

    storage handler •Can now UPDATE, DELETE and INSERT into Hadoop tables, not just SELECT and LOAD DATA Example Impala DDL + DML Commands with Kudu 42 CREATE TABLE `my_first_table` ( `id` BIGINT, `name` STRING ) TBLPROPERTIES( 'storage_handler' = 'com.cloudera.kudu.hive.KuduStorageHandler', 'kudu.table_name' = 'my_first_table', 'kudu.master_addresses' = 'kudu-master.example.com:7051', 'kudu.key_columns' = 'id' ); INSERT INTO my_first_table VALUES (99, "sarah"); INSERT IGNORE INTO my_first_table VALUES (99, "sarah"); UPDATE my_first_table SET name="bob" where id = 3; DELETE FROM my_first_table WHERE id < 3; DELETE c FROM my_second_table c, stock_symbols s WHERE c.name = s.symbol;
  17. Accompanied by Innovations in Underlying Platform 45 Cluster Resource Management

    to
 support mulJ-tenant distributed services In-Memory Distributed Storage,
 to accompany In-Memory Distributed Processing
  18. 50 Copyright © 2015, Oracle and/or its affiliates. All rights

    reserved. | Proprietary ETL engines die circa 2015 – folded into big data Oracle Open World 2015 21 Proprietary ETL is Dead. Apache-based ETL is What’s Next Scripted SQL Stored Procs ODI for Columnar ODI for In-Mem ODI for Exadata ODI for Hive ODI for Pig & Oozie 1990’s Eon of Scripts and PL-SQL Era of SQL E-LT/Pushdown Big Data ETL in Batch Streaming ETL Period of Proprietary Batch ETL Engines Informatica Ascential/IBM Ab Initio Acta/SAP SyncSort 1994 Oracle Data Integrator ODI for Spark ODI for Spark Streaming Warehouse Builder
  19. •By definition there's lots of data in a big data

    system ... so how do you find the data you want? •Google's own internal solution - GOODS ("Google Dataset Search") •Uses crawler to discover new datasets •ML classification routines to infer domain •Data provenance and lineage •Indexes and catalogs 26bn datasets •Other users, vendors also have solutions •Oracle Big Data Discovery •Datameer •Platfora •Cloudera Navigator Google GOODS - Catalog + Search At Google-Scale 53
  20. •Came out if the data science movement, as a way

    to "show workings" •A set of reproducible steps that tell a story about the data •as well as being a better command-line environment for data analysis •One example is Jupyter, evolution of iPython notebook •supports pySpark, Pandas etc •See also Apache Zepplin Web-Based Data Analysis Notebooks 55
  21. Mark Rittman, Oracle ACE Director THE FUTURE OF ANALYTICS, DATA

    INTEGRATION AND BI ON BIG DATA PLATFORMS HADOOP USER GROUP IRELAND (HUG IRL) Dublin, September 2016