AWS Summit Berlin 2012 Talk on Web Data Commons
5 likes10,814 views
This document discusses analyzing large amounts of web data on a startup budget using Amazon Web Services (AWS). It describes how the Common Crawl dataset provides HTML dumps of billions of URLs for free that can be accessed from AWS. Running analysis on this large dataset would be prohibitively expensive without AWS due to computing and storage costs. The document outlines how AWS services like EC2 instances, S3, SQS, and SDB can be used together to parallelize processing across many servers and coordinate the analysis in a cost effective manner of around $1700. Preliminary results of analyzing structured data embedded in web pages are presented.
AWS Summit Berlin 2012 Talk on Web Data Commons
- 1. Large-Scale Analysis of Web Pages − on a Startup Budget? Hannes Mühleisen, Web-Based Systems Group AWS Summit 2012 | Berlin
- 2. Our Starting Point 2
- 3. Our Starting Point • Websites now embed structured data in HTML 2
- 4. Our Starting Point • Websites now embed structured data in HTML • Various Vocabularies possible • schema.org, Open Graph protocol, ... 2
- 5. Our Starting Point • Websites now embed structured data in HTML • Various Vocabularies possible • schema.org, Open Graph protocol, ... • Various Encoding Formats possible • μFormats, RDFa, Microdata 2
- 6. Our Starting Point • Websites now embed structured data in HTML • Various Vocabularies possible • schema.org, Open Graph protocol, ... • Various Encoding Formats possible • μFormats, RDFa, Microdata Question: How are Vocabularies and Formats used? 2
- 7. Web Indices • To answer our question, we need to access to raw Web data. 3
- 8. Web Indices • To answer our question, we need to access to raw Web data. • However, maintaining Web indices is insanely expensive • Re-Crawling, Storage, currently ~50 B pages (Google) 3
- 9. Web Indices • To answer our question, we need to access to raw Web data. • However, maintaining Web indices is insanely expensive • Re-Crawling, Storage, currently ~50 B pages (Google) • Google and Bing have indices, but do not let outsiders in 3
- 10. • Non-Profit Organization 4
- 11. • Non-Profit Organization • Runs crawler and provides HTML dumps 4
- 12. • Non-Profit Organization • Runs crawler and provides HTML dumps • Available data: • Index 02-12: 1.7 B URLs (21 TB) • Index 09/12: 2.8 B URLs (29 TB) 4
- 13. • Non-Profit Organization • Runs crawler and provides HTML dumps • Available data: • Index 02-12: 1.7 B URLs (21 TB) • Index 09/12: 2.8 B URLs (29 TB) • Available on AWS Public Data Sets 4
- 14. Why AWS? • Now that we have a web crawl, how do we run our analysis? • Unpacking and DOM-Parsing on 50 TB? (CPU-heavy!) 5
- 15. Why AWS? • Now that we have a web crawl, how do we run our analysis? • Unpacking and DOM-Parsing on 50 TB? (CPU-heavy!) • Preliminary analysis: 1 GB / hour / CPU possible • 8-CPU Desktop: 8 months • 64-CPU Server: 1 month • 100 8-CPU EC2-Instances: ~ 3 days 5
- 16. Common Crawl Dataset Size
- 17. Common Crawl Dataset Size 1 CPU, 1 h
- 18. Common Crawl Dataset Size 1 CPU, 1 h 1000 € PC, 1 h
- 19. Common Crawl Dataset Size 1 CPU, 1 h 1000 € PC, 1 h 5000 € Server, 1 h
- 20. Common Crawl Dataset Size 1 CPU, 1 h 1000 € PC, 1 h 5000 € Server, 1 h 17 € EC2 Instances, 1 h
- 21. AWS Setup • Data Input: Read Index Splits from S3 7
- 22. AWS Setup • Data Input: Read Index Splits from S3 • Job Coordination: SQS Message Queue 7
- 23. AWS Setup • Data Input: Read Index Splits from S3 • Job Coordination: SQS Message Queue • Workers: 100 EC2 Spot Instances (c1.xlarge, ~0.17 € / h) 7
- 24. AWS Setup • Data Input: Read Index Splits from S3 • Job Coordination: SQS Message Queue • Workers: 100 EC2 Spot Instances (c1.xlarge, ~0.17 € / h) • Result Output: Write to S3 7
- 25. AWS Setup • Data Input: Read Index Splits from S3 • Job Coordination: SQS Message Queue • Workers: 100 EC2 Spot Instances (c1.xlarge, ~0.17 € / h) • Result Output: Write to S3 • Logging: SDB 7
- 26. SQS • Each input file queued in SQS • EC2 Workers take tasks from SQS • Workers read and write S3 buckets 42 ... EC2 42 43 ... R42 R43 ... CC WDC S3
- 27. SQS • Each input file queued in SQS • EC2 Workers take tasks from SQS • Workers read and write S3 buckets 42 ... EC2 42 43 ... R42 R43 ... CC WDC S3
- 28. SQS • Each input file queued in SQS • EC2 Workers take tasks from SQS • Workers read and write S3 buckets 42 ... EC2 42 43 ... R42 R43 ... CC WDC S3
- 29. Results - Types of Data Microdata 02/2012 RDFa 02/2012 Website Structure 23 % 5e+06 RDFa 2009/2010 Microdata 2009/2010 Products, Reviews 19 % Entity Count (log) 5e+05 Movies, Music, ... 15 % 5e+04 Geodata 8 % 5e+03 People, Organizations 7 % 0 50 100 150 200 2012 Microdata Breakdown Type 9
- 30. Results - Types of Data Microdata 02/2012 RDFa 02/2012 Website Structure 23 % 5e+06 RDFa 2009/2010 Microdata 2009/2010 Products, Reviews 19 % Entity Count (log) 5e+05 Movies, Music, ... 15 % 5e+04 Geodata 8 % 5e+03 People, Organizations 7 % 0 50 100 150 200 2012 Microdata Breakdown Type • Available data largely determined by major player support 9
- 31. Results - Types of Data Microdata 02/2012 RDFa 02/2012 Website Structure 23 % 5e+06 RDFa 2009/2010 Microdata 2009/2010 Products, Reviews 19 % Entity Count (log) 5e+05 Movies, Music, ... 15 % 5e+04 Geodata 8 % 5e+03 People, Organizations 7 % 0 50 100 150 200 2012 Microdata Breakdown Type • Available data largely determined by major player support • “If Google consumes it, we will publish it” 9
- 32. Results - Formats 2009/2010 • 4 02−2012 URLs with embedded Data: +6% Percentage of URLs 3 2 1 0 RDFa Microdata geo hcalendar hcard hreview XFN Format 10
- 33. Results - Formats 2009/2010 • 4 02−2012 URLs with embedded Data: +6% Percentage of URLs 3 • Microdata +14% (schema.org?) 2 1 0 RDFa Microdata geo hcalendar hcard hreview XFN Format 10
- 34. Results - Formats 2009/2010 • 4 02−2012 URLs with embedded Data: +6% Percentage of URLs 3 • Microdata +14% (schema.org?) 2 • 1 RDFa +26% (Facebook?) 0 RDFa Microdata geo hcalendar hcard hreview XFN Format 10
- 35. Results - Extracted Data • Extracted data available for download at • www.webdatacommons.org 11
- 36. Results - Extracted Data • Extracted data available for download at • www.webdatacommons.org • Formats: RDF (~90 GB) and CSV Tables for Microformats (!) 11
- 37. Results - Extracted Data • Extracted data available for download at • www.webdatacommons.org • Formats: RDF (~90 GB) and CSV Tables for Microformats (!) • Have a look! 11
- 38. AWS Costs • Ca. 5500 Machine-Hours were required • 1100 € billed by AWS for that 12
- 39. AWS Costs • Ca. 5500 Machine-Hours were required • 1100 € billed by AWS for that • Cost for other services negligible * 12
- 40. AWS Costs • Ca. 5500 Machine-Hours were required • 1100 € billed by AWS for that • Cost for other services negligible * • * At first, we underestimated SDB cost 12
- 41. Takeaways • Web Data Commons now publishes the largest set of structured data from Web pages available 13
- 42. Takeaways • Web Data Commons now publishes the largest set of structured data from Web pages available • Large-Scale Web Analysis now possible with Common Crawl datasets 13
- 43. Takeaways • Web Data Commons now publishes the largest set of structured data from Web pages available • Large-Scale Web Analysis now possible with Common Crawl datasets • AWS great for massive ad-hoc computing power and complexity reduction 13
- 44. Takeaways • Web Data Commons now publishes the largest set of structured data from Web pages available • Large-Scale Web Analysis now possible with Common Crawl datasets • AWS great for massive ad-hoc computing power and complexity reduction • Choose your architecture wisely, test by experiment, for us EMR was too expensive. 13
- 45. Thank You! Questions? Want to hire me? Web Resources: http://webdatacommons.org http://hannes.muehleisen.org