An Introduction To Heritrix Gordon Mohr Chief Technologist

An Introduction To Heritrix Gordon Mohr Chief Technologist, Web Projects Internet Archive

Web Collection • Since 1996 • Over 4 x 1010 resources (URI+time) • Over 400 TB (compressed)

Web Collection: via Alexa • Alexa Internet – Private company – Crawling for IA since 1996 • 2 -month rolling snapshots – Recent: 3 billion URIs, 35 million websites, 20 TB • Crawling software – Sophisticated – Weighted towards popular sites – Proprietary: we only receive the data

Heritrix: Motivations #1 • Deeper, specialized, in-house crawling – Sites of topical interest – Contractual crawls for libraries and governments • US Library of Congress – Elections, current events, government websites • UK Public Records Office, US National Archives – Government websites – Using our own software & machines

Heritrix: Motivations #2 • Open source – Encourage collaboration on features and best practices – Avoid duplication of work, incompatibilities • Archival-quality – Perfect copies – Keep up with changing web – Meet evolving needs of Internet Archive and International Internet Preservation Consortium

Heritrix New Open-source Extensible Web-scale Archival-quality Web crawling software

Heritrix: Use Cases • Broad Crawling – Large, as-much-as-possible • Focused Crawling – Collect specific sites/topics deeply • Continuous Crawling – Revisit changed sites • Experimental Crawling – Novel approaches

Heritrix: Project • Heritrix means heiress • Java, modular • Project website: http: //crawler. archive. org – News, downloads, documentation – Sourceforge: open source hosting site • Source-code control (CVS) • Issue databases • “Lesser” GPL license • Outside contributions

http: //crawler. archive. org

Heritrix: Milestones • Summer 2003: Prototypes created and tested against existing crawlers; requirements collected from IA and IIPC • October 2003 -April 2004: Nordic Web Archive programmers join project, add capabilities • January 2004: First public beta (0. 2. 0) – Used for all in-house crawling since • February & June 2004: Workshops for Heritrix users at national libraries • August 2004: Version 1. 0. 0 released

Heritrix: Architecture • Basic loop: 1. Choose a URI from among all those scheduled 2. Fetch that URI 3. Analyze or archive the results 4. Select discovered URIs of interest, and add to those scheduled 5. Note that the URI is done and repeat • Parallelized across threads (and eventually, machines)

Key components of Heritrix • Scope which URIs should be included (seeds + rules) • Frontier which URIs are done, or waiting to be done (queues and lists/maps) • Processor chains configurable sequential tasks to do to each URI (code modules + configuration)

Heritrix: Architecture

Heritrix: Processor Chains • Prefetch – Ensure conditions are met • Fetch – Network activity (HTTP, DNS, FTP, etc. ) • Extract – Analyze – especially for new URIs • Write – Save archival copy to disk • Postprocess – Feed URIs back to Frontier, update crawler state

Heritrix: Features & Limitations • Other key features: – Web UI console to control & monitor crawl – Very configurable inclusion, exclusion, politeness policies • Limitations: – Requires sophisticated operator – Large crawls hit single-machine limits – No capacity for automatic revisit of changed material • Generally: – Good for focused & experimental crawling use cases; not yet for broad and continuous

Heritrix console

Heritrix settings

Heritrix logs

Heritrix reports

Heritrix: Current Uses • Weekly, Monthly, 6 -monthly, and special one-time crawls • Hundreds to thousands of specific target sites • Over 20 million collected URIs per crawl • Crawls run for 1 -2 weeks

Heritrix: Performance • Not yet stressed, optimized – Current crawls limited by material to crawl and chosen politeness, not our performance • Typical observed rates (actual focused crawls) – 20 -40 URIs/sec (peaking over 60) – 2 -3 Mbps (peaking over 20 Mbps) • Limits imposed by memory usage – Over 10, 000 hosts/over 10 million URIs (512 MB machine, more on larger machines)

Heritrix: Future Plans • Larger scale crawl capacity – Giant focused crawls – Broad whole-web crawls • New protocols & formats • Automate expert operator tasks • Continuous and dynamic crawling – Revisit sites as they change – Dynamically rank sites and URIs

Latest Developments • 1. 2 Release (next week) – Configurable canonicalization • Handles common session-IDs, URI variations – Politeness by IP address – Experimental more memory-efficient Frontier – Bug fixes • 1. 4 Release (January 2004) – Memory robustness – Experimental multi-machine distribution support

The End • Questions?