Syncsort & comScore Big Data Warehouse Meetup Sept 2013
•Download as PPTX, PDF•
2 likes•1,937 views
Mike Brown's CTO of comScore's presentation from the Big Data Warehouse Meetup sponsored by Syncsort Sept 2013 NYC covering how they process over 1.7 Trillion interactions using Hadoop
Recommended
Recommended
More Related Content
What's hot
What's hot (19)
Viewers also liked
Viewers also liked (6)
Similar to Syncsort & comScore Big Data Warehouse Meetup Sept 2013
Similar to Syncsort & comScore Big Data Warehouse Meetup Sept 2013 (20)
Recently uploaded
Recently uploaded (20)
Syncsort & comScore Big Data Warehouse Meetup Sept 2013
- 1. © comScore, Inc. Proprietary. Using Hadoop to Process a Trillion+ Events Michael Brown, CTO | September 23rd, 2013
- 2. © comScore, Inc. Proprietary. 2 comScore is a leading internet technology company that provides Analytics for a Digital World™ NASDAQ SCOR Clients 2,100+ Worldwide Employees 1,000+ Headquarters Reston, Virginia, USA Global Coverage Measurement from 172 Countries; 44 Markets Reported Local Presence 32 Locations in 23 Countries Big Data Over 1 Trillion Digital Interactions Captured Monthly V0113
- 3. © comScore, Inc. Proprietary. Broad Client Base and Deep Expertise Across Key Industries Media Agencies Telecom/Mobile Financial Retail Travel CPG Pharma Technology V0910
- 4. © comScore, Inc. Proprietary. Panel Heat Map
- 5. © comScore, Inc. Proprietary. CENSUS Unified Digital Measurement™ (UDM) Establishes Platform For Panel + Census Data Integration PANEL Unified Digital Measurement (UDM) Patent-Pending Methodology Adopted by 90% of Top 100 U.S. Media Properties Global PERSON Measurement Global DEVICE Measurement V0411
- 6. © comScore, Inc. Proprietary. 6 0 200,000,000,000 400,000,000,000 600,000,000,000 800,000,000,000 1,000,000,000,000 1,200,000,000,000 1,400,000,000,000 1,600,000,000,000 1,800,000,000,000 2,000,000,000,000 Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug 2009 2010 2011 2012 2013 #ofrecords Panel Records Beacon Records Total records collected in August 2013 1,729,895,147,710 Worldwide Tags per Day
- 7. © comScore, Inc. Proprietary. Worldwide UDM™ Penetration December 2012 Penetration Data Europe Austria 87% Belgium 93% Switzerland 89% Germany 92% Denmark 88% Spain 95% Finland 93% France 92% Ireland 90% Italy 90% Netherlands 93% Norway 91% Portugal 92% Sweden 90% United Kingdom 92% Asia Pacific Australia 90% Hong Kong 95% India 92% Japan 82% Malaysia 93% New Zealand 91% Singapore 92% North America Canada 94% United States 91% Latin America Argentina 95% Brazil 96% Chile 94% Colombia 95% Mexico 93% Puerto Rico 92% Middle East & Africa Israel 92% South Africa 78% Percentage of Machines Included in UDM Measurement
- 8. © comScore, Inc. Proprietary. High Level Data Flow Panel Census Custom Code + Delivery
- 9. © comScore, Inc. Proprietary. Our Cluster Production Hadoop Cluster 224 nodes: Mix of Dell 720xd, R710 and R510 servers Each R720xd has (24x1.2TB drives; 64GB RAM; 24 cores) 6300+ total CPUs 13.3TB total memory 4.3PB total disk space Our distro is MapR M5 2.1.3
- 10. © comScore, Inc. Proprietary. The Project: vCE – Validated Campaign Essentials
- 11. © comScore, Inc. Proprietary. 11 vCE provides real-time, cloud- based, on-demand monitoring and optimization of digital advertising campaigns Deep industry penetration 22 of the Top 25 Largest Global Advertisers, representing 89% of global ad dollars, are vCE/CE clients* Includes ALL Top 10 CPG Advertisers* What is vCE? *Source: AdAge 2012 Top 25 Global Advertisers (directly or through their advertising agency) Allstate
- 12. © comScore, Inc. Proprietary. comScore - vCE
- 13. © comScore, Inc. Proprietary. The Problem Statement Calculate the number of events and unique cookies for each reportable campaign element Key take away Data on input will be aggregated daily Need to process all data for 3 months Need to calculate values for every day in the 92 day period spanning all reportable campaign elements
- 14. © comScore, Inc. Proprietary. Structure of the Required Output Client Campaign Population Location Cookie Ct Period 1234 160873284 840 1 863,185 1 1234 160873284 840 1 1,719,738 2 1234 160873284 840 1 2,631,624 3 1234 160873284 840 1 3,572,163 4 1234 160873284 840 1 4,445,508 5 1234 160873284 840 1 5,308,532 6 1234 160873284 840 1 6,032,073 7 1234 160873284 840 1 6,710,645 8 1234 160873284 840 1 7,421,258 9 1234 160873284 840 1 8,154,543 10
- 15. © comScore, Inc. Proprietary. Counting Uniques from a Time Ordered Log File A B C D B A A Major Downsides: Need to keep all key elements in memory. Constrained to one machine for final aggregation.
- 16. © comScore, Inc. Proprietary. First Version Java Map-Reduce application which processes pre-aggregated data from 92 days Map reads the data and emits each cookie as the key of the key value pair All 130B records go though the shuffle Each Reducer will get all the data for a particular campaign sorted by cookie Reducer aggregates the data by grouping key ( Client / Campaign / Population ) and calculates unique cookies for period 1-92 Volume Grew rapidly to the point the daily processing took more than a day
- 17. © comScore, Inc. Proprietary. M/R Data Flow CB Mapper MapperMapperMap Map Map Reduce ReduceReduce BA AC AA BB CC A B C
- 18. © comScore, Inc. Proprietary. Scaling Issue As our volume has grown we have the following stats: Over 500 billion events per month Daily Aggregate 1.5 billion 130 billion aggregate records for 92 days 70K Campaigns Over 50 countries We see 15 billion distinct cookies in a month We only need to output 25 million rows
- 19. © comScore, Inc. Proprietary. Basic Approach Retrospective Processing speed is not scaling to our needs on a sample of the input data Diagnosis Most aggregations could not take significant advantage of combiners. Large shuffles caused poor job performance. In some cases large aggregations ran slower on the Hadoop cluster due to shuffle and skew in data for keys. Diagnosis A new approach is required to reduce the shuffle
- 20. © comScore, Inc. Proprietary. Counting Uniques from a Key Ordered Log File A D B C B A A Major Downsides: Need to sort data in advance. The sort time increases as volume grows.
- 21. © comScore, Inc. Proprietary. Counting Uniques from a Key Ordered Log File
- 22. © comScore, Inc. Proprietary. Counting Uniques from Sharded Key Ordered Log Files
- 23. © comScore, Inc. Proprietary. Solution to reduce the shuffle The Problem: Most aggregations within comScore can not take advantage of combiners, leading to large shuffles and job performance issues The Idea: Partition and sort the data by cookie on a daily basis Create a custom InputFormat to merge daily partitions for monthly aggregations
- 24. © comScore, Inc. Proprietary. Custom Input Format with Map Side Aggregation CB Mapper MapperMapperMap Map Map Reduce ReduceReduce BA AC A B C A B C Combiner Combiner Combiner A B C
- 25. © comScore, Inc. Proprietary. Risks for Partitioning Data locality Custom InputFormat requires reading blocks of the partitioned data over the network This was solved using a feature of the MapR file system. We created volumes and set the chunk size to zero which guarantees that the data written to a volume will stay on one node Map failures might result in long run times Size of the map inputs is no longer set by block size This was solved by creating a large number (10K) of volumes to limit the size of data processed by each mapper
- 26. © comScore, Inc. Proprietary. Partitioning Summary Benefits: A large portion of the aggregation can be completed in the map phase Applications can now take advantage of combiners Shuffles sizes are minimal Results: Took a job from 35 hours to 3 hours with no hardware changes
- 27. © comScore, Inc. Proprietary. DMX & comScore
- 28. © comScore, Inc. Proprietary. DMX use at comScore We use DMX from Syncsort across hundreds of servers for efficient data processing and aggregation. We currently run over 100+ unique jobs every day. With these jobs we process over 150 billion rows of data through DMX! Connect Design Process Accelerate
- 29. © comScore, Inc. Proprietary. Compression w/Sorting Compress Log Files when processing large volumes of log data Several advantages to Sorting Data First: Reduces the size of the data Improves application performance Examples: 1 Hour of one source of our data (313 GB raw, 815 million rows) Standard compression of time ordered data is 93GB (30% of original) Standard compression on a 2 key sorted set is 56GB (18% of original) For one day it saves 800GB When applied to all our sources we save 4.5 TB per day 137 TB per month 412TB per quarter
- 30. © comScore, Inc. Proprietary. TCO with Large Cluster Systems Examine the ability to sort data to reduce disk usage Example: Hadoop cluster that needs to support 100TB of base compressed data Hypothetical Configurations @ 75% disk utilization: Replication Factor of 3 using 1.2 TB drives R710 (6x drives, JBOD); requires 26 servers R510 (12x drives JBOD); requires 52 servers R720xd (24x drives JBOD); requires 13 servers
- 31. © comScore, Inc. Proprietary. Useful Factoids Visit www.comscoredatamine.com or follow @datagems for the latest gems. Colorful, bite-sized graphical representations of the best discoveries we unearth.
- 32. © comScore, Inc. Proprietary. Thank You! Michael Brown CTO comScore, Inc. mbrown@comscore.com
Editor's Notes
- Key MessagecomScore is a global internet technology company providing customers with Analytics for a Digital WorldSupporting Talking PointsFounded in 1999, comScore is best known as the gold standard for measuring digital activity, including website visitation, search, video, social, digital advertisingcomScore’s data and technologies are well-established crucial components in measuring and analyzing the rapidly evolving digital world, and are widely deployed at a broad range of publishers, advertising agencies, advertisers, retailers and telecom operators, both in the US and internationally
- comScore leverages DMExpress from SyncSort across hundreds of our servers to allow us to efficiently process our data.A generic design pattern for us is to sort the input data based on the column that we will be counting uniques. Counting uniques is one of the more costly measures to calculate in a system. By sorting the data in advance, you only need to see if the prior value has changed from the current value and increment a counter.This approach has let us implement aggregation systems that can process over 50 GB of data with 357 million rows in less than an hour on a Dell R710 2U server.